FRANK, Circuit Judge (dissenting).

1. I have no respect for the humorless self-righteous sort of person who has a firm conviction that always he alone, of the entire regiment, is in step. Accordingly, when all my colleagues (whom I consider among the ablest of judges) repeatedly arrive at a certain conclusion, my sense of humor usually downs my doubts and nudges me into acquiescence. But on the subject of "harmless error" in criminal trials, I find myself, because of the deep seriousness of the matter, unable to follow that course. I am emboldened to persist in my disagreement by the fact that three times in the past few years the Supreme Court has reversed my colleagues for their views on that issue. See Berger v. United States, 1935, 295 U.S. 78, 84-89, 55 S.Ct. 629, 79 L.Ed. 1314, reversing 2 Cir., 73 F.2d 278; Bruno v. United States, 1939, 308 U.S. 287, 293, 294, 60 S.Ct. 198, 84 L.Ed. 257, reversing 2 Cir., 105 F.2d 921; Bollenbach v. United States, 66 S.Ct. 402, reversing, 2 Cir., 147 F.2d 199.

2. As crucial facts have been omitted from or glossed over in the majority opinion, I must narrate them:

Some of the defendants are Italian-born and others of Italian descent. One of them, Antonelli, testified in a decided Italian-American jargon. American soldiers were still fighting in Italy when the case was tried. The closing arguments to the jury occurred on June 8, 1944, just after our Army's invasion of Normandy had begun. In those circumstances (and perhaps bearing in mind an ancient observation, "If you want to excite prejudice you must do so at the close, so that the jurors may more easily remember what you said.") government counsel concluded his summation with this final sentence: "I cherish an overwhelming confidence * * * in the belief that * * * you will render a verdict of which you can be proudly justified in the presence of your fellowmen, those here at home who labor and have labored unceasingly in an honest effort to manufacture munitions of war as well as those of us beyond the seas who look to us for the things they need to sustain them in their hour of extreme sacrifice." Counsel for the defendants thereupon objected to these improper remarks.

At a minimum, so the Supreme Court tells us, the trial judge, in such circumstances, "without waiting for an objection," should have promptly employed a "stern rebuke and repressive measures." But here the judge was neither prompt nor stern, and uttered no rebuke. My colleagues are entirely mistaken when they say that he acted "immediately after the summation," applied "extensive corrective measures," gave a "rebuke" which "was swift and sure," and "deflated immediately" the words of the prosecutor. For it was not until the day after the summation — on the next morning when the jury reconvened after a recess — that he took any steps to correct the error. Then, for the first time (and, contrary to my colleagues' statement, not with defendants' approval) he gave but a mild cautionary instruction, devoid of anything resembling a rebuke, stern or otherwise. Thus at least some twelve hours had intervened, so that, during the jurors' overnight reflections on the case, this vicious appeal to their prejudices had had a chance to soak in, before the judge mildly counselled the jurors on the subject — and this in a case which (as I shall try to show) was by no means "strong." The inflammatory and highly prejudicial character of such a gross stimulation of such patriotic sentiments when a criminal trial is conducted in wartime has been universally recognized. See, e.g., Viereck v. United States, 318 U.S. 236, 247-278, 63 S.Ct. 561, 87 L.Ed. 734; People v. Levan, 295 N.Y. 26, 35, 36, 64 N.E.2d 341; and see the cases dealing with criminal trials during World War I. Never before has any court said what my colleagues say here — that such deadly remarks were "no more than an admonition to the jurors to observe their oath of office and thus have the satisfaction of duty well done." Nor heretofore has any court deemed such remarks harmless because compressed into a single sentence, for experience teaches that a poisonous suggestion of that kind needs no elaboration. As I shall show below, the courts have frequently held that this type of error cannot be eradicated by a cautionary instruction or even by an immediate severe censuring of offending counsel. For these reasons I think this error was so markedly harmful that it requires reversal for a new and fair trial. On that ground I dissent.

A "strong" case has been defined as one in which the evidence of guilt is "overwhelming," Berger v. United States, 295 U.S. 78 at 89, 55 S.Ct. 629, 79 L.Ed. 1314. In an Appendix to this dissent I have stated in some detail some of the significant evidence to which my colleagues have but sketchily and quite inadequately referred; it will there be seen, not only that the jury could reasonably have found defendants not guilty, (i.e., could reasonably have concluded that the defective work was due not to fraud but to the inexperience and incompetence of Antonelli, the other defendants, and the employees) but also that among the government's important witnesses were self-confessed accomplices and — more important — that there was considerable testimony, much of it uncontradicted, that government agents had practiced intimidation of witnesses. I therefore say that the government's case was not "strong." The courts recognize the unreliability of accomplices' testimony and have often declared that a party's intimidation of witnesses is a "badge of the weakness" of his case. Wigmore, Evidence § 278, says: "It has always been understood — the inference, indeed, is one of the simplest in human experience — that a party's falsehood or other fraud in the preparation and presentation of his cause, his fabrication or suppression of evidence by bribery or spoliation, and all similar conduct, is receivable against him as an indication of his consciousness that his case is a weak or unfounded one, and from that consciousness may be inferred the fact itself of the cause's lack of truth and merit. The inference thus does not apply itself necessarily to any specific fact in the cause, but operates, indefinitely though strongly, against the whole mass of alleged facts constituting his cause * * * As the general principle applies in common to all these forms of conduct, it is not necessary, nor is it usually possible, to discriminate the precedents that apply it in one or another form. Roughly classifying them, they admit all forms of personal falsification by the party in the course of the litigation; fabrication or manufacture of evidence, by forgery, bribery, subornation, and the like; suppression of evidence, by intimidation, eloignment, or concealment of witnesses or material objects * * *" Usually in criminal cases that rule has been applied to defendants. But it applies to the government as well. United States v. Graham, 2 Cir., 102 F.2d 436, 442.

3. As, of course, no court knows what influenced a particular jury's verdict of guilt in any particular case, there are three possible alternative positions with respect to "harmless error." The first is that every error is harmful; no court today takes that position. The second — which is that of the Supreme Court, of most of the circuit courts and of the English courts — may be summarized thus: An error (except as to "formal matters" i.e., those which involve the "mere etiquette of trials * * * and minutiae of procedure") is presumed to be prejudicial (i.e., improperly to have induced the verdict); but this presumption is rebutted if the evidence is so "strong" that no sensible jury, had there been no error, would conceivably have acquitted, as for instance where the defendant in his testimony in effect admits his guilt. In other words, when there is substantial error the appellate court will not affirm merely because the jury, if it believed the government's witnesses and disbelieved defendant's, could reasonably have inferred that defendant was guilty; it will affirm only if, had no error occurred, the jury could not reasonably have reached a different verdict, and will reverse "if, upon any conceivable construction of anything in the testimony, it would have been possible for a reasonable man to have reached any other verdict than the one returned by the jury."

As I have said elsewhere, this rule does not work automatically, for some limited conjecture is unavoidable; but this rule does severely restrict the area of conjecture. It means that the judges do not themselves decide the issue of guilt, do not consider the credibility of the witnesses, do not invade the jury's province.

Necessarily, an upper court which does not employ either of these rules employs — either explicitly or tacitly — one which involves its own determination of the defendant's guilt or innocence, although the testimony is in conflict and the court has not heard and seen the witnesses. Such a court will hold an error "harmless" if its judges believe, from their study of the printed record, that defendant is guilty, regardless of whether those judges are convinced that the evidence is such that, absent the error, a reasonable jury which heard the witnesses would indubitably have rendered a verdict against defendant. In a court which proceeds on that doctrine of harmless error, the judges decide against the defendant, not by affirming a jury's verdict on the record which was presented to the jury, but on a markedly different record — one from which the error has been elided. The judges of such a court impliedly say in effect: "To be sure, the matter erroneously injected into the case when it was before the jury may have diverted the jury from a consideration of the properly admitted evidence and may have persuaded the jurors to find defendant guilty without regard to whether they believed the government's witnesses and disbelieved defendant's. But no one can tell that that is true, and we will not disturb a conviction on mere guesswork. So we use a different approach: We have read the record, and, while we did not observe the witnesses, and therefore lack the accepted means of determining credibility, we believe the testimony of those witnesses called by the government. We do not know, of course, whether, had we been present at the trial, we would so have believed; but we must do the best we can in dealing with such a conjectural subject. Accordingly, we hold that defendant is guilty, and consequently conclude that it is unimportant whether in actual fact (which is unknowable) the improper matter, rather than the evidence, induced the jury's verdict." The use of that doctrine means that the appellate judges are returning their own verdict, independent of (although preceded by) the jury's, and founded on a record other than that which the jury considered; for the judges are able to and do disregard the improper matter, but it is impossible to know that the jury did. Since the appellate judges' verdict rests on printed testimony, the result is a juryless "trial by affidavits." Such a doctrine, which does indeed compel extensive judicial guesswork, is inescapable for a court which rejects both the other two positions described above.

It is this third doctrine which (with certain exceptions, noted in the Appendix hereto, but not pertinent here) has been heretofore adopted by my colleagues and which they apply here.

Their position is illuminated by their decision in a recent case, United States v. Mitchell, 2 Cir., 137 F.2d 1006, 1010, in which the testimony was in conflict, and in which they held harmless unfairness towards the defendant on the part of the trial judge. There my colleagues refused to reverse because they concluded that the case was "strong," although that conclusion turned on my colleagues' opinion as to the credibility of the witnesses. On rehearing — 138 F.2d 831 — a revised transcript of the trial court's record disclosed the inaccuracy of important "facts" on which my colleagues had previously based their views as to the "strength" of the case. My colleagues nevertheless adhered to their former decision, although no one reading the revised version of the facts (set forth in 138 F.2d at pages 832, 833) could possibly say that a jury would have been at all unreasonable had it acquitted. Obviously when my colleagues said that that case was "strong," they meant no more than that, without themselves observing the witnesses, they believed defendant guilty.

And so in the case at bar where, again, the testimony is in conflict: Referring to one of defendants' contentions, my colleagues say that "the jury was justified in concluding otherwise." They state that the verdict "was amply supported by the evidence," since "questions of credibility are for the jury." They conclude that "the evidence was as strong and persuasive as it is ever likely to be in a hotly disputed case," and that, "supported by the natural inferences to be drawn," the verdict was "thoroughly justified." Such expressions would be entirely appropriate if there had been no error in the course of the trial and if the issue were whether there was sufficient competent evidence to support the verdict, for then the question would be whether the court could say that no jury could rationally have found defendants guilty. But such expressions are not appropriate where (as here) the court is called on to determine whether, in a "hotly contested case," turning on "questions of credibility," it may properly hold that, if no error had occurred, no jury could rationally have found the defendants innocent.

To restate the point: In this, as in other "harmless error" cases, my colleagues have lost sight of a distinction — which I think must be made — between the function of a reviewing court when (1) the issue is whether the verdict is against the weight of the evidence and (2) when the issue is whether an error (such as, e.g., erroneous admission of evidence or misconduct of counsel) is "harmless." Because they have repeatedly ignored this distinction, I think it desirable (at the cost of some repetition) to dwell on it: In a case where no prejudicial evidence has been received and where no misconduct of government counsel has occurred, if a defendant asks an upper court to reverse his conviction on the sole ground that the record evidence is such that no reasonable jury could have found him guilty, the judges are compelled to resort to a sort of fiction: Since they were absent from the trial and since the credibility of the witnesses is a question for the jury, the judges, perforce, must assume it to be true (whether or not it is true in fact) that all the testimony pointing to guilt was given by honest, reliable, credible witnesses and that all other testimony was not. That assumption (a fiction, an "as if") in such a case is unavoidable; therefore necessary; therefore proper. Be it noted, however, that, in such circumstances, the judges at least have before them, although in printed form, the identical matter — neither more nor less — which had been presented to the jury. But that assumption (fiction) is not necessary and is therefore, I think, improper, when the record contains prejudicial evidence improperly admitted, or remarks of counsel improperly made. For the issue on appeal is then not whether, making that assumption and on the very same record which was before the jury, the jury reasonably found against defendant. If, in a case like this, the judges conclude that defendant is guilty, they reach that conclusion by ignoring important matter which had been presented to the jury and which may well have led to its verdict. In those circumstances, the judges decide against defendant, not by affirming a jury's verdict on the record which the jury had before it, but on a strikingly different record.

My colleagues' doctrine runs directly contrary to the rule enunciated in an opinion delivered by Mr. Justice Miller where (answering a contention that an error in admitting evidence did no harm because "There is enough found in the record to show that the verdict was right, if it had been excluded"), he said for the Court: "The case must be such that this court is not called on to decide upon the preponderance of evidence that the verdict was right, notwithstanding the error complained of." That was a civil case; it goes without saying that the rule should be at least as favorable to a defendant in a criminal suit.

My colleagues, however, assert that such a rule is today outmoded: Recently, in United States v. Liss, 2 Cir., 137 F.2d 995, 999, they said that there is a "modern disposition to assume that an error has been harmless * * *." As I have pointed out elsewhere, that view is opposed to the current rulings of the Supreme Court and of the other circuits. I think that Bollenbach v. United States, supra, confirms my position. There the Supreme Court said that the question is not whether the appellate judges have the belief, "engendered by the dead record," that defendants are guilty, "but whether guilt has been found by a jury according to the procedure and standards appropriate for criminal trials in the federal courts." [66 S. Ct. 406.] When the Supreme Court in Bollenbach's case declared that it is wrong to presume "all errors to be `harmless' if only the appellate court is left without doubt that one who claims its protective process is, after all, guilty," it was not speaking at large, but was engaged in reversing this court; that serves, I believe, to show that the Supreme Court thinks, as I do, that my colleagues, in cases where the testimony is in conflict, erroneously rest their decisions as to harmless error on their own belief in a defendant's guilt or innocence.

Decisions of the English courts, recently cited with approval by the Supreme Court in Bollenbach's case, are instructive. In each of the following cases thus cited, the Court was requested to affirm a conviction pursuant to a statute (§ 4 of the Criminal Appeals Act of 1907) much like our own "harmless error" statute. In Maxwell v. The Director of Public Prosecutions, [1935] A.C. 309, defendant had been improperly asked about a previous prosecution for another crime of which he had been acquitted; the court in reversing, said, "It is impossible to deny that the evidence was strong against him," as defendant had made "very damaging admissions," but it was not certain what the jury would have decided, had the question not been asked. The court concluded thus: "If in any case the evidence against a prisoner (other than that which is inadmissible) is very strong and is abundant to justify a jury in convicting, it may well seem unfortunate that a guilty man should go free because some rule of evidence has been infringed by the prosecutor. But it must be remembered that the whole policy of English criminal law has been to see that as against the prisoner every rule in his favour is observed and that no rule is broken so as to prejudice the chance of the jury fairly trying the true issues. The sanction for the observance of the rules of evidence in criminal cases is that, if they are broken in any case, the conviction may be quashed * * * It is often better that one guilty man should escape than that the general rule evolved by the dictates of justice for the conduct of criminal prosecutions should be disregarded and discredited." In Rex v. Dyson, [1908] 2 K.B. 454, where an improper instruction was given, the court reversed although it felt that the jury, if it had been given a proper instruction, "would in all probability have found the defendant guilty," saying that the statute (as to "substantial miscarriage of justice") was "intended to apply to a case in which the evidence is such that the jury must have found the prisoner guilty if they had been properly directed." The following English cases, in each of which the trial court erred in its directions to the jury, are also pertinent: In Woolmington v. The Director of Public Prosecutions, [1935] A.C. 462, 482-483, the conclusion was, "We cannot say that if the jury had been properly directed they would inevitably have come to the same conclusion." In Rex v. Lewis, 31 Cox Cr. L.Cas., 19, 22-23 (Ct. of Crim.App., 1937), the test was said to be whether "the conclusion is not to be resisted, that the jury, properly directed, would certainly have arrived at the same conclusion." In Rex v. Haddy, 1944, 1 K.B. 442-446, the Court used the test that "no reasonable jury properly directed would, or could, have come to any other conclusion than that to which they did come." In Stirland v. Director of Public Prosecutions, [1944] A.C. 315, 321, the test used was whether a "reasonable jury, after a proper summing up, could have failed to convict," or would "without doubt convict." Our Supreme Court's formulation is virtually the same.

The reason for this rule is obvious: "It is seldom possible with even moderate competence to conjecture solely from perusal of a written or printed record whether or not a defendant is guilty. As the judges of an appeal court have not heard or seen the witnesses, they have no reasonably adequate way of judging whether any of the witnesses lied or — because of unconscious bias or faulty memory — testified with inadvertent inaccuracy."

Most (not all) of the precautionary and exclusionary rules have derived from, or have been perpetuated because of, the recognition of the untrained capacity of jurors; for that reason, departures from those rules are usually not error when a trial judge sits without a jury (e.g., reception of inadmissible evidence when there is other competent evidence to sustain the judgment). But if in a jury trial in a criminal suit (a) there has been a violation of any of these rules — designed to keep from the jurors matter which might improperly influence them against the defendant — and (b) the verdict is adverse to him, and (c) the evidence is not such that, had that rule not been violated, a reasonable jury would unquestionably have reached the same verdict, then, unless the verdict is set aside, the very purpose of the rule is frustrated. The point is that, if such a violation is held harmless, the rule has been abandoned, and the jury has been subjected to influences recognized, in the judicial adoption of the rule, as unfair to the defendant. In sum, unless the rule is senseless and should never have been adopted, the defendant has not received a fair trial.

4. Applying the usual "harmless error" doctrine, the courts generally hold that improper remarks (or other similar misconduct) of counsel will be deemed to have induced the verdict (Berger v. United States, 295 U.S. 78, 85-89, 55 S.Ct. 629, 79 L.Ed. 1314 ) and to require reversal. For such remarks may affect the jury even more than erroneously admitted evidence. Close students of the subject, such as Morgan, tell us that today, unfortunately, a jury trial usually is "a game in which the contestants are not the litigants but the lawyers." An experienced trial lawyer writes: "It is a well recognized fact that in most cases the jury `tries' the lawyers rather than the clients * * * The personality of the lawyer is constantly before the jury and he gradually absorbs the client's cause to such an extent that unconsciously in the minds of the jury it becomes the lawyer's cause." And the courts have said that the words of the government's lawyer are likely to be exceptionally impressive, since he is an official.

Exception (as usual) is made when the evidence against the defendant is "strong," i.e., the evidence of guilt is "overwhelming," so that it is all but impossible to believe that any sensible jury would have acquitted even if no improper remarks had been made; in such a case, no court will reverse, no matter how seriously the government counsel misbehaved.

United States v. Socony-Vacuum Oil Co., 310 U.S. 150, 60 S.Ct. 811, 84 L.Ed. 1129, is a typical exceptional case. There the Court (310 U.S. at page 177, 60 S.Ct. 811, 824, 84 L.Ed. 1129) pointed out that evidence sustaining the verdict consisted "almost entirely" of "unequivocal testimony or undisputed contents of exhibits, only occasionally on the irresistible inferences [drawn] from those facts"; accordingly the Court — citing with approval but distinguishing Berger v. United States supra, — said (310 U.S. at page 239, 60 S.Ct. at page 852, 84 L.Ed. 1129) that "this was not a weak case," and that therefore the improper remarks of government counsel did not constitute prejudicial error requiring reversal. Also illustrative of the kind of case in which misconduct of the United States Attorney will properly not induce reversal is Robbins v. United States, 9 Cir., 229 F. 987, 988, where the Court said the verdict "rested upon evidence so clear and convincing that the jury could not have determined otherwise than as they did," and the jury was therefore "compelled to reach" that verdict, despite the misconduct. That, and Horning v. District of California, 254 U.S. 135, 41 S.Ct. 53, 65 L.Ed. 185, and Socony-Vacuum, supra, show what is meant by a "strong" case.

Very different is a case where the testimony is sharply in conflict, so that the credibility of the witnesses is crucial, and the jury could reasonably have found the defendant either innocent or guilty. In those circumstances, although, absent error, the verdict will not be set aside as unsupported by the evidence, nevertheless such a case is not "strong" and therefore a conviction will be reversed because of egregious misconduct by government counsel. And surely, the evidence of guilt is not "overwhelming" where, as here, not only is the testimony in sharp conflict, but the government's case depends in considerable part on testimony of accomplices (see Berger's case, 295 U.S. at page 89, 55 S.Ct. 629, 79 L.Ed. 1314) and there is undisputed evidence of witness intimidation (as shown in the Appendix hereto).

There is no room for the suggestion that Socony-Vacuum overruled Berger's case; for the latter was cited with approval, since Socony-Vacuum, in Viereck v. United States, 318 U.S. at page 248, 63 S.Ct. 561, 87 L.Ed. 734. Moreover, the Court still applies its usual "strong case" test when it is claimed that an error did no harm; so in Glasser v. United States, 315 U.S. 60, 67, 62 S.Ct. 457, 463, 86 L.Ed. 680, decided after Socony-Vacuum, the Court, in reversing a conviction, characterized the government's case as not "strong," adding, "Then error, which under some circumstances would not be ground for reversal, cannot be brushed aside as immaterial since there is a real chance that it might have provided the slight impetus which swung the scales toward guilt."

5. True, it is generally held that error consisting of misconduct of counsel will be deemed to have been cured if the trial court promptly gave the offending counsel a "stern rebuke," and immediately cautioned the jury to disregard the misconduct. But when (a) counsel's remarks are of an unusually inflammatory character (or he has otherwise flagrantly misbehaved) and (b) the evidence of guilt is not "overwhelming," then almost everywhere in criminal cases (except in this circuit) and virtually everywhere in civil cases, it has been held that neither a cautionary instruction nor any other action by the trial judge, by way of rebuke or otherwise, suffices to purge the error, but that he should grant a mistrial or a new trial, and that, if he does not do so, the judgment must be reversed.

See, e.g., Pharr v. United States, 6 Cir., 48 F.2d 767, 770, 771; Towbin v. United States, 10 Cir., 93 F.2d 861, 868; Volkmor v. United States, 6 Cir., 13 F.2d 594, 595; Skuy v. United States, 8 Cir., 261 F. 317, 319, 320; Robinson v. United States, 8 Cir., 32 F.2d 505, 508, 66 A.L.R. 468; Maytag v. Cummins, 8 Cir., 260 F. 74, 82, 16 A.L.R. 712; Latham v. United States, 5 Cir., 226 F. 420, 425; Frisby v. United States, 35 App.D.C. 513, 520; Vaughan v. Magee, 3 Cir., 218 F. 630, 631, 632; James Stewart Co. v. Newby, 4 Cir., 266 F. 287, 295, 296; Beck v. Wings Field, Inc., 3 Cir., 122 F.2d 114, 117; Pierce v. United States, 6 Cir., 86 F.2d 949, 952, 953; People v. Levan, 295 N.Y. 26, 36, 64 N.E.2d 341; People v. Fielding, 158 N.Y. 542, 53 N.E. 497, 46 L.R.A. 641, 70 Am.St.Rep. 495; People v. White, 365 Ill. 499, 6 N.E.2d 1015, 1012; Lickliter v. Commonwealth, 249 Ky. 95, 60 S.W.2d 355, 357; Cassemus v. State, 16 Ala. App. 61, 75 So. 267, 268; cf. Waldron v. Waldron, 156 U.S. 361, 383, 384, 15 S.Ct. 383, 39 L.Ed. 453; Throckmorton v. Holt, 180 U.S. 552, 567, 21 S.Ct. 474, 45 L.Ed. 663; Lockhart v. United States, 9 Cir., 35 F.2d 905, 907; Middleton v. United States, 8 Cir., 49 F.2d 538, 540.

This rule was well stated a few months ago (November 29, 1945) in People v. Levan, 295 N.Y. 26, 34-36, 64 N.E.2d 341, 345, where the court reversed a conviction for murder; there the district attorney told the jury: "Now, this defendant appears here in this court room in an Army uniform. I don't suppose the fact that he wears this uniform means a thing, as far as any effect on your judgment, gentlemen. * * * There are many men wearing this uniform and doing their duty in other parts of the world beside Harlem and New York City." These remarks were not appreciably longer than those of government counsel in the case at bar. But the highest court of New York (citing Viereck's and Berger's cases, supra) ruled as follows: "It should be said that in the case now before us the jury was told by the court `* * * you are not going to convict this defendant because there are other men on the other side being shot, or serving their country. You are not to convict him on that theory at all.' We cannot say, however, with assurance that the jury did disregard the clear implications from the emphasis placed by the prosecuting attorney upon the damning fact that the defendant had absented himself without leave from his military duties in a time of war. The virus thus implanted in the minds of the jury is not so easily extracted." In Berger v. United States, supra, the Court concluded by citing as "especially apposite," People v. Wells, 100 Cal. 459, 34 P. 1078, 1079, 1080. In the Wells case the court quoted with approval from People v. Ah Len, 92 Cal. 282, 28 P. 286, 27 Am.St.Rep. 103, where a conviction was reversed for grave improprieties by government counsel although "the trial court warned the jury specially on the subject."

Indeed, the judge's cautionary instruction may do more harm than good: It may emphasize the jury's awareness of the censured remark — as in the story, by Mark Twain, of the boy told to stand in the corner and not think of a white elephant.

6. Particularly with respect to improper remarks of counsel, my colleagues in criminal cases do not follow the usual rule. They argue in effect, that, as the matter is inherently guessy, it is impossible ever to be sure that improper remarks of counsel influenced the verdict; for that reason, this court does not, in criminal cases, reverse for such an error. The rule in this court appears, I think, from the following: Beginning with 1934, and prior to the instant case, the question has arisen five times in this circuit. In United States v. Berger, 1934, 2 Cir., 73 F.2d 278, 281, this court said that the United States Attorney had seriously misbehaved but that "it would be extravagant now to reverse the conviction on" that account, "for it is fantastic to suppose that it substantially determined the outcome," adding that "to-day, when mere possibilities do not interest us as they did our forerunners, we demand more tangible evidence that damage has been done." The Supreme Court reversed that decision, because (as previously noted) it said the case was "weak." Berger v. United States, 1935, 295 U.S. 78, 85-89, 55 S.Ct. 629, 79 L. Ed. 1314. Thereafter, in four cases, this court, while citing the Supreme Court's decision in Berger's case, affirmed convictions, notwithstanding what it conceded to be grave misconduct by government counsel. See United States v. Dubrin, 1937, 2 Cir., 93 F.2d 499, 506; United States v. Lotsch, 1939, 2 Cir., 102 F.2d 35, 37; United States v. Weiss, 1939, 2 Cir., 103 F.2d 348, 355; United States v. Buckner, 1940, 2 Cir., 108 F.2d 921, 928, 929. In other words, since the time it was reversed in Berger's case, this court, in every instance where a United States Attorney has seriously misbehaved, has found that the case against the defendant was so "strong" that the Berger doctrine was inapplicable. It might be thought that, in each of these instances, there was, in fact, "overwhelming" evidence of guilt. But I believe there is a different explanation of at least some of those decisions: My colleagues simply do not accept the usual definition of a "strong" case.

I think this appears from United States v. Mitchell, discussed above. In like manner, in United States v. Buckner, 2 Cir., 108 F.2d 921, this court refused to reverse for counsel's shocking improprieties because it held (page 928) the "proof is strong and convincing", although it said (page 926) as to some of the defendants that unlawful intent "could clearly have been inferred" — not at all that no other reasonable inference could have been drawn; and it said (page 930) as to one of the defendants, that "the jury was amply justified in so concluding * * * that [he] knew of and assisted in the conspiracy" — not at all that no other reasonable conclusion was possible. See how different were these facts from those in Socony-Vacuum; and compare the reasoning of this court in the Buckner case with that in Volkmor v. United States, 6 Cir., 13 F.2d 594, 595, where the circuit court reversed because of remarks of government counsel (which, under the admonition of the trial court, he withdrew) although the circuit court held the evidence sufficient to sustain the verdict, saying, "From these facts, with such explanation as defendant offered, the jury might or might not have inferred a fraudulent intent." Patently, in Buckner's case, as in United States v. Mitchell, supra, my colleagues' affirmance rested on their own belief in defendants' guilt, despite the conflicting testimony. And so, as I have already shown, in the instant case.

My colleagues mistakenly state, as to the "men overseas" remarks of government counsel, that the trial court "promised to take care of the matter in its charge," and that this course was "apparently completely acceptable to the defendants." As will be seen from the colloquy quoted in the note, the trial judge made no such promise, and defendants did not therefore indicate that such a course would be acceptable. True, having promptly and forcefully objected to the "men overseas" argument of counsel, defendants' lawyers did not go further and ask for a mistrial or object to the judge's charge in that respect after it was given. But it is revelatory that their failure to do so has been treated by my colleagues as a waiver of defendants' right to assign error now. For, even if defendants had not so much as mentioned the matter in the trial court, my colleagues would not hesitate to note it on appeal, if they regarded the error as grave enough to be a ground of reversal. In Pierce v. United States, 6 Cir., 86 F.2d 949, 953, 954, the court rejected the contention "that the improper conduct of the prosecuting attorneys was waived by failure of the defendants to move for a mistrial." Judge Simons said: "Above and beyond all technical procedural rules, designed to preserve the rights of litigants, is the [public's] maintenance of the nation's courts as fair and impartial forums where neither bias nor prejudice rules, and appeals to passion find no place, though the government itself be there a litigant. * * * Where such paramount considerations are involved, procedural niceties will not preclude a court from correcting error." See New York C.R. Co. v. Johnson, 279 U.S. 310, 318, 49 S.Ct. 300, 73 L.Ed. 706. We have so held as to other kinds of error; see e.g., United States v. Trypuc, 2 Cir., 136 F.2d 900, 902; United States v. Haug, 2 Cir., 150 F.2d 911; and our Rule 10. That my colleagues refuse so to hold here is a fairly obvious sign that they hold that improper remarks of government counsel are not sufficient to compel reversal when my colleagues believe a defendant guilty.

Notwithstanding, as previously noted, that two of the government's important witnesses were self-confessed accomplices and that there was uncontradicted testimony of witness intimidation by the government, my colleagues argue that the case here is "strong". Their argument, spelled out, runs thus: (1) As the verdict was against the defendants, it must be assumed (according to my colleagues) that, although the testimony was in conflict, the jury disbelieved the defendants' witnesses (including their testimony about intimidation), believed the government's witnesses, and, accordingly, drew inferences from the defects in the products to the effect that they showed fraud on the part of the defendants; (2) therefore (so my colleagues seem to imply) we must similarly disbelieve the defendants' witnesses, believe the government's, and draw the same inferences; (3) from this it follows (my colleagues reason) that the case was "strong"; (4) consequently (say my colleagues), we must conclude that the case, not being "weak," the improper remarks of government counsel did not influence the jury. Plainly, such an argument proves too much; it wholly destroys the ruling in Berger's case and similar cases: Since the question of the effect on the jury of government counsel's remarks can never arise unless there is a verdict of guilt, the result must be that, if my colleagues are correct (i.e., that a verdict of guilty is invariably to be taken as demonstrating that the evidence of guilt was so overwhelming that the misconduct did not influence the jury), in no case can such conduct be reversible error — for then every case in which a defendant appeals is a "strong" case. Surely, correct reasoning runs the other way: As the evidence of witness intimidation and the use of accomplices' testimony disclose a recognition by the government of the "weakness" of the government's case, it should be assumed that the case is not "strong" and that the verdict against the defendants resulted from the misconduct (i.e., that the evidence is not such that, absent the misconduct, a reasonable jury would indubitably have found defendant guilty.)

I think it plain, then, that, because of their unique conception of "harmless error" and their interpretation of a "strong case", my colleagues do not follow the rule in Berger's case, but, remaining renitent, render it mere lip service: Just as they refuse to reverse for erroneously admitted evidence when, despite conflicting testimony, they believe a defendant guilty, so they do likewise as to counsel's improper remarks in a criminal suit.

Strangely enough, my colleagues have taken a different position as to misconduct of counsel in civil litigation. In Brown v. Walter, 2 Cir., 62 F.2d 798, 799, 800, the lawyer for the successful plaintiff, in an automobile collision case, injected into the record the fact that the nominal defendant was insured by an insurance company. Although the trial judge cautioned the jury not to heed that fact this court reversed, citing cases — including James Stewart Co. v. Newby, 4 Cir., 266 F. 287, 295, 296, and Brooke v. Croson, 61 App.D.C. 159, 58 F.2d 885 — to effect that "no caution would serve to cure" such an error. Surely, if that rule is to be invoked to protect the pocketbook of an insurance company, it should be invoked in the instant case to protect natural persons from being sent to jail unjustly.

A jury trial, at best, is chancy. "Mr. Prejudice and Miss Sympathy are the names of witnesses whose testimony is not recorded, but must nevertheless be reckoned with * * *"; and most jurors have no trained capacity for doing so. A keen observer has said that "next to perjury, prejudice is the main cause of miscarriages of justice." If government counsel in a criminal suit is allowed to inflame the jurors by irrelevantly arousing their deepest prejudices, the jury may become in his hands a lethal weapon directed against defendants who may be innocent. He should not be permitted to summon that thirteenth juror, prejudice. Law suits, do what we will, are hazardous: A missing witness, a lost document — these and numerous other fortuitous factors may result in a man's losing his life, liberty or property unjustly. When the government puts a citizen to the hazards of a criminal jury trial, a government attorney should not be allowed to increase those hazards unfairly. When, as here, such an attorney has done so, I, as a government servant, am unwilling to approve the result. I think it is our duty to give these defendants another trial.

7. The majority opinion here contains an implication that only a super-sentimental or excessively soft-minded judge, inadequately aware of the need to safeguard the public from criminals, would favor reversal of a conviction because of the sort of error present here. Perhaps I am such a mushy-souled sentimentalist. But no such charge can be leveled at the many judges who, in reversing convictions, have employed the rule I think applicable here. (Perhaps they share with G.K. Chesterton a disdain for "that miserable fear of being sentimental, which is the meanest of all the modern terrors.") I refer to Chief Justice Stone, Mr. Justice Sutherland, and to circuit judges such as, e.g., Simons, McDermott, and Sanborn. In Miller v. Territory of Oklahoma, 8 Cir., 149 F. 330, 339, 9 Ann.Cas. 389, Judge Philips said: "The foregoing incident strikingly illustrates where the responsibility for the miscarriage of justice in criminal prosecutions should sometimes be placed, instead of imputing the reversal of convictions by the appellate courts to what is popularly termed mere technicalities.' The zeal, unrestrained by legal barriers, of some prosecuting attorneys, tempts them to an insistence upon the admission of incompetent evidence, or getting before the jury some extraneous fact supposed to be helpful in securing a verdict of guilty, where they have prestige enough to induce the trial court to give them latitude. When the error is exposed on appeal, it is met by the stereotyped argument that it is not apparent it in any wise influenced the minds of the jury. The reply the law makes to such suggestion is: that, after injecting it into the case to influence the jury, the prosecutor ought not to be heard to say, after he has secured a conviction, it was harmless. As the appellate court has not insight into the deliberations of the jury room, the presumption is to be indulged, in favor of the liberty of the citizen, that whatever the prosecutor, against the protest of the defendant, has laid before the jury, helped to make up the weight of the prosecution which resulted in the verdict of guilty." And see Judge Sanborn's opinion in Skuy v. United States, 8 Cir., 261 F. 316, 319, 320, where the court reversed for improper remarks of counsel, despite the trial judge's cautionary instruction. The English cases, discussed above, are also pertinent here; for English judges are not proverbially chicken-hearted, nor has administration of criminal justice in England been notoriously lax.

8. Judges who accept what I consider the correct view of harmless error do not demand perfection in trials, or engage in fly-specking scrutiny of trial records. Also, in condemning appeals to prejudice, they have in mind the different meanings of the word "prejudice": Every society has its fundamental "value judgments," which constitute its established predilections, ideals, preconceptions — "prejudices" in that sense of the word. To the extent, then, that a government lawyer appeals to those cherished values, so far as they are relevant to the case in hand, he acts appropriately. So, too, when he urges the jury to discern the hidden motives of the defendant's witnesses. "Impartiality is not gullibility. Disinterestedness does not mean child-like innocence." But a government lawyer acts unfairly when he arouses jury prejudices that are irrelevant and distracting.

9. This court has several times used vigorous language in denouncing government counsel for such conduct as that of the United States Attorney here. But, each time, it has said that, nevertheless, it would not reverse. Such an attitude of helpless piety is, I think, undesirable. It means actual condonation of counsel's alleged offense, coupled with verbal disapprobation. If we continue to do nothing practical to prevent such conduct, we should cease to disapprove it. For otherwise it will be as if we declared in effect, "Government attorneys, without fear of reversal, may say just about what they please in addressing juries, for our rules on the subject are pretend-rules. If prosecutors win verdicts as a result of `disapproved' remarks, we will not deprive them of their victories; we will merely go through the form of expressing displeasure. The deprecatory words we use in our opinions on such occasions are purely ceremonial." Government counsel, employing such tactics, are the kind who, eager to win victories, will gladly pay the small price of a ritualistic verbal spanking. The practice of this court — recalling the bitter tear shed by the Walrus as he ate the oysters — breeds a deplorably cynical attitude towards the judiciary.

On the other hand, a reversal in a case like this might well serve as a deterrent: If it became known that misconduct of a United States Attorney had caused the public the expense of a new trial, his resultant unpopularity might tend to make him subsequently live up to professional standards of courtroom decency. If this court really meant business about such behavior as that of government counsel in the case at bar, if it actually considered such behavior reprehensible, it would, at a minimum, announce that if, in any future case any government lawyer should thus conduct himself, it would deprive him of the right to practice in this court and would recommend that he be removed from his office as a representative of our government.

This is no light matter. In its Report on Lawlessness in Law Enforcement, the National Committee on Law Observance and Enforcement said, in 1931, of unfairness in prosecutions (including "the various forms of misconduct by prosecutors * * * in the courtroom," such as unfair and inflammatory comments and appeals to prejudice): "First, these unfair practices are a type of lawless enforcement of law which is especially liable to create resentment against law and government, because they are committed by district attorneys * * * — the very officials most definitely responsible for law observance. Moreover, these abuses usually occur in the publicity of a courtroom. They are not hidden away and subject to denial like the third degree. They are witnessed by spectators and may be recorded by the press, so that many members of the public may be revolted by the oppressive conduct of men chiefly responsible for the administration of justice. Such resentment easily engenders the dangerous feeling that a fair trial has been denied because the defendant belongs to an unpopular group and that for members of such a group justice through the courts is not to be expected * * * Not to be overlooked is the effect of unfairness upon the accused. Even if he is guilty there are degrees of criminality which he may not yet have reached. It may still be possible to accomplish his readjustment to society, but hardly so if he feels deeply and justly that society in the person of its chief representatives has behaved tyrannically and brutally. The natural effect of this motion is to alienate him still further from the community and make him regard his criminal associates as the only men who treat him decently. In consequence he may leave prison a bitter enemy of society, more willing than before to continue a criminal career. His resentment will be shared by his family and friends. The result of the unfairness upon these persons and upon the public will be a decrease in respect for law, which is a main factor in assuring its observance * * * Thirdly, and perhaps most seriously, unfair practices may result in the conviction of the innocent."

A legal system is not what it says, but what it does. Our "criminal law," then, cannot be described accurately in terms merely of substantive prohibitions; the description must also include the methods by which those prohibitions operate in practice — must include, therefore, not the substantive and procedural rules as they appear in words but as they actually work, or, as Llewellyn puts it, "the net operation of the whole official set-up, taken as a whole," for it "is that net operation — it is the substantive rule only as it trickles through the screen of action — which counts in life." With opinions like that in the instant case — in which my colleagues go so far as to express a fear that reversal for a new trial would be "deadening to the morale" of the prosecutor — the "official set-up," as sanctioned by this court, will surely include a disregard by government lawyers of precepts against rabble-rousing jury speeches.

I have spelled out this dissent at what may appear to be unseemly length because it deals with a subject I consider of the gravest importance for this reason: Order is a necessary condition of social existence; to attain that order, a society makes laws, the infraction of which it punishes; one of the functions of courts is to direct such punishment. But our kind of society deems it an essential principle that no court shall direct that any man be punished except after a fair trial. As no one can prove that a society rejecting that principle cannot attain order, some skeptics sneer at that principle, call it super-refined nonsense. It is, however, part of the American faith that, without that principle, a society would be inadequately civilized. That faith our courts, I think, should vindicate by a jealous insistence that trials should not only appear to be but should actually be fair. In the vale of human perplexities, perfection in that endeavor is impossible. But recognition of that impossibility should not deter us from approaching as near the ideal of such fairness as we can. And I think this court needlessly falls far short of it, if it affirms a conviction of defendants, obtained in marked violation of the rules governing fair trials, merely because the judges of this court believe those defendants guilty. If it does so, it is, I think, helping to undermine a basic tenet of the American faith. That seems to me to be dangerous in these days when America is seeking to induce the world to accept its conception of civilization as a pattern of world order. Perhaps my sense of humor has indeed deserted me, and I indulge in exaggeration; but I think not: The courts alone can neither create moral principles nor tear them down, but they can be among the vital agencies which either preserve or corrode them.

Lawyers may talk rhapsodically of JUSTICE. They may, in Bar Association meetings, hymn the pre-eminent virtues of "our Lady of the Common Law," prostrate themselves devotedly before the miracle of the common law's protection of human liberties. But, in the last analysis, there is only one practical way to test puddings: If, again and again in concrete instances, courts unnecessarily take the chance of having innocent men sent to jail or put to death by the government because they have been found guilty by juries persuaded by unfair appeals to improper prejudices, then the praises of our legal system will be but beautiful verbal garlands concealing ugly practices we have not the courage, or have grown too callous, to contemplate.

Some judges disavow responsibility for such ugliness, asserting that the judiciary has no concern with whether or not the existing legal machinery yields injustice. They agree with Lord Sumner that judges "are not now free in the twentieth century to administer the vague jurisprudence which is sometimes attractively styled `justice as between man and man.'" Happily, that sort of "dispassionateness" does not represent the current judicial ideal. Most judges today acknowledge that administering justice constitutes at least part of their official obligations. That acknowledgment, however, remains worse than empty unless it translates itself into action.

Of course, no human contrivances for getting at the fact of guilt or innocence can be infallible. Even if the greatest care be used, some guiltless men are bound to be punished. That risk seems to many reflective thinkers greater than it should be, because we ask casually selected groups of twelve persons, most of them untrained in the difficult art of fact-finding, to make the decisions. But, since our society continues to prize the jury as a fact-finding instrument, we are committed to taking that extra risk. For that very reason, I submit, judges should be extraordinarily vigilant to prevent the unfortunate consequences of carelessly utilizing that risky instrument.

Something like our modern jury was employed by the ancient Greeks. One of the wisest of those Greeks gave his fellows some advice we would do well to follow: "It is not right to pervert the jurymen by moving them to anger or envy or pity — one might as well warp a carpenter's rule before using it."

Appendix

I. Evidence Showing that the Government's case was not "strong."

1. The following evidence goes to show that the defective work may well have been due, not to fraud, but to the inexperience and incompetence of Antonelli, the other defendants and the employees.

Antonelli, the sole stockholder of the company, was born in Italy and came to this country at the age of 21, some thirty years ago. He is uneducated, illiterate, unable to speak English correctly. His business had been that of making fireworks. Before 1941, he had had in his employ only a few persons except for a short period each year, when sometimes he had as many as twenty. He and the other defendants had had no experience in the making of products by pressure or in mass production until he received the government contracts, involved in this litigation, which called for the rapid assembling of several millions of bombs and required the employment of about 350 persons. Antonelli was being urged by the government to speed up his work.

Colonel Zanetti, one of the government's witnesses, testified as follows: The machines used in connection with the government contracts did not supply the required pressure automatically. The uniformity of the pressure depended entirely upon the sense of touch or sight of the girls operating the machines. "It was left up to the girl." Experience taught that uniform compression could have been obtained by an automatic process; absent that process, the necessary result could be obtained only by constant inspection. Zanetti said that the type of labor available to Antonelli was not reliable.

2. The following testimony concerning witness intimidation is of marked significance as showing recognition by the government of the weakness of its case:

Costanza testified that Kiefer and other F.B.I. agents had given him to understand that "they would make things easy for him" if he "cooperated with him," and that, "while they could not promise to get me off completely, they would see to it that everything would go easy for me." Beatrice De Filippo, an employee, testified that when interviewed by F.B.I. agents they had assured her that if she would testify in a certain way she would have nothing to fear. John DiRitis testified that he made an engagement with three employees (Lucy Sigilone, Johanna Arone, and Theresa Molinari) to come to the office of one of the defense counsel so that such counsel could interview them in connection with the preparation of the defense; but that when he (DiRitis) called for these witnesses, they refused to see the defense counsel because they had been told by Kiefer, an F.B.I. agent, not to do so. After this testimony was given, Kiefer who had previously testified, was recalled to the stand by the government, but was not asked any questions concerning it, and in no way contradicted it.

Johanna Arone, an employee and a government witness, testified on cross-examination, that she had been questioned by representatives of the F.B.I., and that fear of harm that might come to her influenced her to testify against some of the defendants.

Two of the chief government witnesses were Bianchi and Pitio, self-confessed accomplices in the alleged conspiracy. Pitio had a criminal record. In his testimony he admitted that he had told a prospective witness that "if he knew what was good for him, he would get out of this lawsuit and not testify for the defense." Alice LaBrutto testified that Pitio had told her not to testify: her testimony was not contradicted.

I fail to comprehend why my colleagues refer to the testimony, above described, as "trivial stuff" that "deserved the complete ignoring which the government gave it."

II. Exceptions made by this Court to its "Harmless Error" Rule.

In fairness to my colleagues, I must say, as I have said elsewhere, that to their working rule of harmless error, they make these exceptions: (a) when important evidence has been excluded; (b) when the trial judge has given an erroneous charge as to the substantive law; (c) sometimes when the trial judge has been extravagantly unfair; (d) when defendant has been deprived of a basic constitutional right (e.g., to be represented by counsel); (e) occasionally, when the sentence is excessive. But for other types of substantial error — such, e.g., as erroneous admission of evidence or such as that here involved — they refuse to reverse if they think defendant guilty, although the testimony is in conflict.

III. A Judge's Doubts about the Wisdom of the Jury System are not Incompatible with His Obligation to Ensure Fair Jury Trials.

It has been suggested that a judge (like me) who shares the doubts about the wisdom of the jury system is inconsistent if he urges that the courts be vigilant in preserving the jury's function. I do not understand that criticism. It is the sworn duty of judges to enforce many statutes they may deem unwise. And so, when on the bench, our private views concerning the desirability of the jury system are "as irrelevant as our attitudes towards bimetallism or the transmigration of souls." Consequently, as long as jury trials are guaranteed by constitutional or statutory provisions, it is the obligation of every judge, no matter what he thinks of such trials, to see that they are fairly conducted and that the jury's province is not invaded. That does not mean that a judge may not freely express his skepticism about the system, may not seek to bring about constitutional and statutory changes which will avoid or reduce what he considers its unfortunate results as it now operates.