MARSHALL, J., dissenting 473 U. S. Thus, while I believe that some review for harmlessness is in order, I disagree with the Court's standard, even were it merely a standard for review and not a definition of “materiality." First, I see no significant difference for truthseeking purposes between the Giglio situation and this one; for the same reasons I believe the result must therefore be the same here as in Giglio, see supra, at 691-692, I also believe the standard for reversal should be the same. The defendant's entitlement to a new trial ought to be no different in the two cases, and the burden he faces on appeal should also be the same. Giglio remains the law for a class of cases, and I reaffirm my belief that the same standard applies to this case as well. See Agurs, supra, at 119-120 (MARSHALL, J., dissenting). Second, only a strict appellate standard, which places on the prosecutor a burden to defend his decisions, will remove the incentive to gamble on a finding of harmlessness. Any lesser standard, and especially one in which the defendant bears the burden of proof, provides the prosecutor with ample room to withhold favorable evidence, and provides a reviewing court with a simple means to affirm whenever in its view the correct result was reached. This is especially true given the speculative nature of retrospective review: "The appellate court's review of 'what might have been' is extremely difficult in the context of an adversarial system. Evidence is not introduced in a vacuum; rather, it is built upon. The absence of certain evidence may thus affect the usefulness, and hence the use, of other evidence to which defense counsel does have access. Indeed, the absence of a piece of evidence may affect the entire trial strategy of defense counsel." Capra, supra, at 412. As a consequence, the appellate court no less than the prosecutor must substitute its judgment for that of the trier of fact under an inherently slippery test. Given such factors as a reviewing court's natural inclination to affirm a judgment 667 MARSHALL, J., dissenting that appears "correct" and that court's obvious inability to know what a jury ever will do, only a strict and narrow test that places the burden of proof on the prosecutor will begin to prevent affirmances in cases in which the withheld evidence might have had an impact. Even under the most protective standard of review, however, courts must be careful to focus on the nature of the evidence that was not made available to the defendant and not simply on the quantity of the evidence against the defendant separate from the withheld evidence. Otherwise, as the Court today acknowledges, the reviewing court risks overlooking the fact that a failure to disclose has a direct effect on the entire course of trial. Without doubt, defense counsel develops his trial strategy based on the available evidence. A missing piece of information may well preclude the attorney from pursuing a strategy that potentially would be effective. His client might consequently be convicted even though nondisclosed information might have offered an additional or alternative defense, if not pure exculpation. Under such circumstances, a reviewing court must be sure not to focus on the amount of evidence supporting the verdict to determine whether the trier of fact reasonably would reach the same conclusion. Instead, the court must decide whether the prosecution has shown beyond a reasonable doubt that the new evidence, if disclosed and developed by reasonably competent counsel, would not have affected the outcome of trial." 'For example, in United States ex rel. Butler v. Maroney, 319 F. 2d 622 (CA3 1963), the defendant was convicted of first-degree murder. Trial counsel based his defense on temporary insanity at the time of the murder. During trial, testimony suggested that the shooting might have been the accidental result of a struggle, but defense counsel did not develop that defense. It later turned out that an eyewitness to the shooting had given police a statement that the victim and Butler had struggled prior to the murder. If defense counsel had known before trial what the eyewitness had seen, he might have relied on an additional defense, and he might have emphasized the struggle. See Note, The Prosecutor's Constitutional MARSHALL, J., dissenting 473 U. S. In this case, it is readily apparent that the undisclosed information would have had an impact on the defense presented at trial, and perhaps on the judgment. Counsel for Bagley argued to the trial judge that the Government's two key witnesses had fabricated their accounts of the drug distributions, but the trial judge rejected the argument for lack of any evidence of motive. See supra, at 690. These key witnesses, it turned out, were each to receive monetary rewards whose size was contingent on the usefulness of their assistance. These rewards "served only to strengthen any incentive to testify falsely in order to secure a conviction." Ante, at 683. To my mind, no more need be said; this non Duty to Reveal Evidence to the Defendant, 74 Yale L. J. 136, 145 (1964). Unless the same information already was known to counsel before trial, the failure to disclose evidence of that kind simply cannot be harmless because reasonably competent counsel might have utilized it to yield a different outcome. No matter how overwhelming the evidence that Butler committed the murder, he had a right to go before a trier of fact and present his best available defense. Similarly, in Ashley v. Texas, 319 F. 2d 80 (CA5), cert. denied, 375 U. S. 931 (1963), the defendant was sentenced to death for murder. The prosecutor disclosed to the defense a psychiatrist's report indicating that the defendant was sane, but he failed to disclose the reports of a psychiatrist and a psychologist indicating that the defendant was insane. The nondisclosed information did not relate to the trial defense of self-defense. But the failure to disclose the evidence clearly prevented defense counsel from developing the possibly dispositive defense that he might have developed through further psychiatric examinations and presentation at trial. The nondisclosed evidence obviously threw off the entire course of trial preparation, and a new trial was in order. In such a case, there simply is no need to consider-in light of the evidence that actually was presented and the quantity of evidence to support the verdict returned-the possible effect of the information on the particular jury that heard the case. Indeed, to make such an evaluation would be to substitute the reviewing court's judgment of the facts, including the previously undisclosed evidence, for that of the jury, and to do so without the benefit of competent counsel's development of the information. See also Field, Assessing the Harmlessness of Federal Constitutional Error-A Process in Need of a Rationale, 125 U. Pa. L. Rev. 15 (1976) (discussing application of harmless-error test). 667 STEVENS, J., dissenting disclosure could not have been harmless. I would affirm the judgment of the Court of Appeals. JUSTICE STEVENS, dissenting. This case involves a straightforward application of the rule announced in Brady v. Maryland, 373 U. S. 83 (1963), a case involving nondisclosure of material evidence by the prosecution in response to a specific request from the defense. I agree that the Court of Appeals misdescribed that rule, see ante, at 674-678, but I respectfully dissent from the Court's unwarranted decision to rewrite the rule itself. As the Court correctly notes at the outset of its opinion, ante, at 669, the holding in Brady was that "the suppression by the prosecution of evidence favorable to an accused upon request violates due process where the evidence is material either to guilt or to punishment." 373 U. S., at 87. We noted in United States v. Agurs, 427 U. S. 97, 103 (1976), that the rule of Brady arguably might apply in three different situations involving the discovery, after trial, of evidence that had been known prior to trial to the prosecution but not to the defense. Our holding in Agurs was that the Brady rule applies in two of the situations, but not in the third. The two situations in which the rule applies are those demonstrating the prosecution's knowing use of perjured testimony, exemplified by Mooney v. Holohan, 294 U. S. 103 (1935), and the prosecution's suppression of favorable evidence specifically requested by the defendant, exemplified by Brady itself. In both situations, the prosecution's deliberate nondisclosure constitutes constitutional error-the conviction must be set aside if the suppressed or perjured evidence was "material" and there was "any reasonable likelihood" that it "could have affected" the outcome of the trial. 427 U. S., at 103.1 See Brady, supra, at 88 ("would tend to exculpate"); 'I do not agree with the Court's reference to the "constitutional error, if any, in this case," see ante, at 678 (emphasis added), because I believe a violation of the Brady rule is by definition constitutional error. Cf. United STEVENS, J., dissenting 473 U. S. accord, United States v. Valenzuela-Bernal, 458 U. S. 858, 874 (1982) ("reasonable likelihood"); Giglio v. United States, 405 U. S. 150, 154 (1972) (“reasonable likelihood"); Napue v. Illinois, 360 U. S. 264, 272 (1959) (“may have had an effect on the outcome"). The combination of willful prosecutorial suppression of evidence and, "more importantly," the potential "corruption of the truth-seeking function of the trial process" requires that result. 427 U. S., at 104, 106.2 In Brady, the suppressed confession was inadmissible as to guilt and "could not have affected the outcome" on that issue. 427 U. S., at 106. However, the evidence "could have affected Brady's punishment," and was, therefore, “material on the latter issue but not on the former." Ibid. Material States v. Agurs, 427 U. S., at 112 (rejecting rule making "every nondisclosure . . . automatic error" outside the Brady specific request or perjury contexts). As written, the Brady rule states that the Due Process Clause is violated when favorable evidence is not turned over "upon request" and "the evidence is material either to guilt or punishment." Brady v. Maryland, 373 U. S., at 87. As JUSTICE MARSHALL's explication of the record in this case demonstrates, ante, at 685-692, the suppressed evidence here was not only favorable to Bagley, but also unquestionably material to the issue of his guilt or innocence. The two witnesses who had signed the undisclosed "Contract[s] for Purchase of Information" were the only trial witnesses as to the two distribution counts on which Bagley was convicted. On cross-examination defense counsel attempted to undercut the witnesses' credibility, obviously a central issue, but had little factual basis for so doing. When defense counsel suggested a lack of credibility during final argument in the bench trial, the trial judge demurred, because "I really did not get the impression at all that either one or both of these men were trying at least in court here to make a case against the defendant." A finding that evidence showing that the witnesses in fact had a "direct, personal stake in respondent's conviction," ante, at 683, was nevertheless not "material" would be egregiously erroneous under any standard. 2 "A prosecution that withholds evidence on demand of an accused which, if made available, would tend to exculpate him or reduce the penalty helps shape a trial that bears heavily on the defendant. That casts the prosecutor in the role of an architect of a proceeding that does not comport with standards of justice. . . ." Brady, supra, at 87-88. |