Opinion

JUSTICE O'CONNOR, dissenting.

I have no dispute with the Court's observation that "collateral review is different from direct review." Ante, at 633. Just as the federal courts may decline to adjudicate certain issues of federal law on habeas because of prudential concerns, see Withrow v. Williams, post, at 686; post, at 699-700 (O'CONNOR, J., concurring in part and dissenting in part), so too may they resolve specific claims on habeas using different and more lenient standards than those applicable on direct review, see, e. g., Teague v. Lane, 489 U. S. 288, 299-310 (1989) (habeas claims adjudicated under the law prevailing at time conviction became final and not on the basis of intervening changes of law). But decisions concerning the Great Writ "warrant restraint," Withrow, post, at 700 (O'CONNOR, J., concurring in part and dissenting in part), for we ought not take lightly alteration of that "'fundamental safeguard against unlawful custody,'" post, at 697-698 (quoting Fay v. Noia, 372 U. S. 391, 449 (1963) (Harlan, J., dissenting)).

In my view, restraint should control our decision today.

The issue before us is not whether we should remove from the cognizance of the federal courts on habeas a discrete prophylactic rule unrelated to the truthfinding function of trial, as was the case in Stone v. Powell, 428 U. S. 465 (1976), and more recently in Withrow v. Williams, post, p. 680. Rather, we are asked to alter a standard that not only finds application in virtually every case of error but that also may be critical to our faith in the reliability of the criminal process. Because I am not convinced that the principles governing the exercise of our habeas powers-federalism, finality, and fairness-counsel against applying Chapman's harmlesserror standard on collateral review, I would adhere to our former practice of applying it to cases on habeas and direct review alike. See ante, at 630. I therefore respectfully dissent.

The Court begins its analysis with the nature of the constitutional violation asserted, ante, at 628-630, and appropriately so. We long have recognized that the exercise of the federal courts' habeas powers is governed by equitable principles. Fay v. Noia, supra, at 438; Withrow, post, at 699-700 (O'CONNOR, J., concurring in part and dissenting in part). And the nature of the right at issue is an important equitable consideration. When a prisoner asserts the violation of a core constitutional privilege critical to the reliability of the criminal process, he has a strong claim that fairness favors review; but if the infringement concerns only a prophylactic rule, divorced from the criminal trial's truthfinding function, the prisoner's claim to the equities rests on far shakier ground. Thus, in Withrow v. Williams, this Court declined to bar relitigation of Miranda claims on habeas because Miranda is connected to the Fifth Amendment and the Fifth Amendment, in turn, serves the interests of reliability. Withrow, post, at 691-692. I dissented because I believe that Miranda is a prophylactic rule that actually impedes the truthseeking function of criminal trials. Withrow, post, at 700, 701-708. See also Stone v. Powell, supra, at 486, 490 (precluding review of exclusionary rule violations in part because the rule is judicially fashioned and interferes with the truthfinding function of trial).

Petitioner in this case alleged a violation of Doyle v. Ohio, 426 U. S. 610 (1976), an error the Court accurately characterizes as constitutional trial error. Ante, at 629-630. But the Court's holding today, it turns out, has nothing to do with Doyle error at all. Instead, the Court announces that the harmless-error standard of Chapman v. California, 386 U. S. 18, 24 (1967), which requires the prosecution to prove constitutional error harmless beyond a reasonable doubt, no longer applies to any trial error asserted on habeas, whether it is a Doyle error or not. In Chapman's place, the Court substitutes the less rigorous standard of Kotteakos v. United States, 328 U. S. 750, 776 (1946). Ante, at 638.

A repudiation of the application of Chapman to all trial errors asserted on habeas should be justified, if at all, based on the nature of the Chapman rule itself. Yet, as JUSTICE WHITE observes, ante, at 645 (dissenting opinion), one searches the majority opinion in vain for a discussion of the basis for Chapman's harmless-error standard. We are left to speculate whether Chapman is the product of constitutional command or a judicial construct that may overprotect constitutional rights. More important, the majority entirely fails to discuss the effect of the Chapman rule. If there is a unifying theme to this Court's habeas jurisprudence, it is that the ultimate equity on the prisoner's side-the possibility that an error may have caused the conviction of an actually innocent person-is sufficient by itself to permit plenary review of the prisoner's federal claim. Withrow, post, at 700 (O'CONNOR, J., concurring in part and dissenting in part) (citing cases). Whatever the source of the Chapman standard, the equities may favor its application on habeas if it substantially promotes the central goal of the criminal justice system-accurate determinations of guilt and innocence. See Withrow, post, at 705-706 (reasoning that, although Miranda may be a prophylactic rule, the fact that it is not "divorced" from the truthfinding function of trial weighs in favor of its application on habeas); Teague, supra, at 313 (if absence of procedure seriously diminishes likelihood of accurate conviction, new rule requiring such procedure may be retroactively applied on habeas).

In my view, the harmless-error standard often will be inextricably intertwined with the interest of reliability. By now it goes without saying that harmless-error review is of almost universal application; there are few errors that may not be forgiven as harmless. Arizona v. Fulminante, 499 U. S. 279, 306-307 (1991). For example, we have recognized that a defendant's right to confront the witnesses against him is central to the truthfinding function of the criminal trial. See, e. g., Maryland v. Craig, 497 U. S. 836, 845-847 (1990); Ohio v. Roberts, 448 U. S. 56, 65 (1980); Mattox v. United States, 156 U. S. 237, 242-243 (1895); see also 3 W. Blackstone, Commentaries 373-374 (1768). But Confrontation Clause violations are subject to harmless-error review nonetheless. See Coy v. Iowa, 487 U. S. 1012, 1021-1022 (1988). When such an error is detected, the harmless-error standard is crucial to our faith in the accuracy of the outcome: The absence of full adversary testing, for example, cannot help but erode our confidence in a verdict; a jury easily may be misled by such an omission. Proof of harmlessness beyond a reasonable doubt, however, sufficiently restores confidence in the verdict's reliability that the conviction may stand despite the potentially accuracy impairing error. Such proof demonstrates that, even though the error had the potential to induce the jury to err, in fact there is no reasonable possibility that it did. Rather, we are confident beyond a reasonable doubt that the error had no influence on the jury's judgment at all. Cf. In re Winship, 397 U. S. 358, 363-364 (1970) (proof of guilt beyond a reasonable doubt indispensable to community's respect and confidence in criminal process).

At least where errors bearing on accuracy are at issue, I am not persuaded that the Kotteakos standard offers an adequate assurance of reliability. Under the Court's holding today, federal courts on habeas are barred from offering relief unless the error" 'had substantial and injurious effect or influence in determining the jury's verdict.'" Ante, at 637 (quoting Kotteakos, supra, at 776). By tolerating a greater probability that an error with the potential to undermine verdict accuracy was harmful, the Court increases the likelihood that a conviction will be preserved despite an error that actually affected the reliability of the trial. Of course, the Constitution does not require that every conceivable precau tion in favor of reliability be taken; and certainly 28 U. S. C. § 2254 does not impose such an obligation on its own. Indeed, I agree with the Court that habeas relief under § 2254 is reserved for those prisoners "whom society has 'grievously wronged.'" Ante, at 637. But prisoners who may have been convicted mistakenly because of constitutional trial error have suffered a grievous wrong and ought not be required to bear the greater risk of uncertainty the Court now imposes upon them. Instead, where constitutional error may have affected the accuracy of the verdict, on habeas we should insist on such proof as will restore our faith in the verdict's accuracy to a reasonable certainty. Adherence to the standard enunciated in Chapman requires no more; and the equities require no less.

To be sure, the harmless-error inquiry will not always bear on reliability. If the trial error being reviewed for harmlessness is not itself related to the interest of accuracy, neither is the harmless-error standard. Accordingly, in theory it would be neither illogical nor grudging to reserve Chapman for errors related to the accuracy of the verdict, applying Kotteakos' more lenient rule whenever the error is of a type that does not impair confidence in the trial's result. But the Court draws no such distinction. On the contrary, it holds Kotteakos applicable to all trial errors, whether related to reliability or not. The Court does offer a glimmer of hope by reserving in a footnote the possibility of an exception: Chapman may remain applicable, it suggests, in some "unusual" cases. But the Court's description of those cases suggests that its potential exception would be both exceedingly narrow and unrelated to reliability concerns. See ante, at 638, n. 9 (reserving the "possibility that in an unusual case, a deliberate and especially egregious error of the trial type" or error "combined with a pattern of prosecutorial misconduct, might so infect the integrity of the proceeding as to warrant the grant of habeas relief, even it did not substantially influence the jury's verdict"). But even if the Court's holding were limited to errors divorced from reliability concerns, the decision nevertheless would be unwise from the standpoint of judicial administration. Like JUSTICE WHITE, I do not believe we should turn our habeas jurisprudence into a "patchwork" of rules and exceptions without strong justification. Ante, at 649 (dissenting opinion). The interest of efficiency, always relevant to the scope of habeas relief, see, e. g., Stone, 428 U. S., at 491, n. 31; Withrow, post, at 693-694; post, at 708-713 (O'CONNOR, J., concurring in part and dissenting in part), favors simplification of legal inquiries, not their multiplication. A rule requiring the courts to distinguish between errors that affect accuracy and those that do not, however, would open up a whole new frontier for litigation and decision. In each case, the litigants would brief and federal judges would be required to decide whether the particular error asserted relates to accuracy. Given the number of constitutional rules we have recognized and the virtually limitless ways in which they might be transgressed, I cannot imagine that the benefits brought by such litigation could outweigh the costs it would impose.

In fact, even on its own terms the Court's decision buys the federal courts a lot of trouble. From here on out, prisoners undoubtedly will litigate-and judges will be forced to decide-whether each error somehow might be wedged into the narrow potential exception the Court mentions in a footnote today. Moreover, since the Court only mentions the possibility of an exception, all concerned must also address whether the exception exists at all. I see little justification for imposing these novel and potentially difficult questions on our already overburdened justice system.

Nor does the majority demonstrate that the Kotteakos standard will ease the burden of conducting harmless-error review in those cases to which it does apply. Indeed, as JusTICE STEVENS demonstrates in his concurrence, Kotteakos is unlikely to lighten the load of the federal judiciary at all. The courts still must review the entire record in search of conceivable ways the error may have influenced the jury; they still must conduct their review de novo; and they still must decide whether they have sufficient confidence that the verdict would have remained unchanged even if the error had not occurred. See ante, at 641-642. The only thing the Court alters today is the degree of confidence that suffices. But Kotteakos' threshold is no more precise than Chapman's; each requires an exercise of judicial judgment that cannot be captured by the naked words of verbal formulae. Kotteakos, it is true, is somewhat more lenient; it will permit more errors to pass uncorrected. But that simply reduces the number of cases in which relief will be granted. It does not decrease the burden of identifying those cases that warrant relief.

Finally, the majority considers the costs of habeas review generally. Ante, at 637. Once again, I agree that those costs-the effect on finality, the infringement on state sovereignty, and the social cost of requiring retrial, sometimes years after trial and at a time when a new trial has become difficult or impossible-are appropriate considerations. See Withrow, post, at 703-704 (O'CONNOR, J., concurring in part and dissenting in part); see also post, at 686-687, 708-709; Stone, supra, at 489-491. But the Court does not explain how those costs set the harmless-error inquiry apart from any other question presented on habeas; such costs are inevitable whenever relief is awarded. Unless we are to accept the proposition that denying relief whenever possible is an unalloyed good, the costs the Court identifies cannot by themselves justify the lowering of standards announced today. The majority, of course, does not contend otherwise; instead, it adheres to our traditional approach of distinguishing between those claims that are worthy of habeas relief and those that, for prudential and equitable reasons, are not. Nonetheless, it seems to me that the Court's decision cuts too broadly and deeply to comport with the equitable and remedial nature of the habeas writ; it is neither justified nor justifiable from the standpoint of fairness or judicial efficiency. Because I would remand the case to the Court of Appeals for application of Chapman's more demanding harmless-error standard, I respectfully dissent.