with whom Justice Marshall joins, concurring in part and dissenting in part.
In this case the Court of Appeals issued an opinion reversing the convictions of the respondents. 660 F. 2d 301 (CA7 1981). Most of the opinion consists of a discussion of the facts. Id., at 301-303. In its brief legal analysis, the court relied on its decision in United States v. Buege, 578 F. 2d 187 (1978), to find that the prosecutor had violated the respondents’ Fifth Amendment rights by commenting on their failure to testify. 660 F. 2d., at 303. The court declined to apply the harmless-error doctrine to this violation. The court stated that an application of the doctrine “would impermissi-bly compromise the clear constitutional violation of [respondents’] Fifth Amendment rights.” Ibid.
On its face, the Court of Appeals’ opinion adopts a rule of automatic reversal for improper prosecutorial comment on a defendant’s failure to testify. Such a rule was rejected by this Court in Chapman v. California, 386 U. S. 18, 22 (1967). The respondents argue that the Court of Appeals’ decision to disregard Chapman was justified as an exercise of the court’s supervisory powers. Brief for Respondents 15-36. I would *520reject this argument on the ground that the Court of Appeals did not invoke its supervisory powers or provide any explanation of why this might be an appropriate case for the exercise of such powers. In order to rely on its supervisory powers to reverse a conviction a court of appeals should be required, at the least, to invoke them expressly.1 In view of the Court of Appeals’ refusal to apply the harmless-error doctrine announced in Chapman and its failure to offer any reasons to justify its refusal, I would vacate the court’s decision and remand the case for application of the harmless-error test announced by Chapman and a determination of whether the error in this case was harmless beyond a reasonable doubt.2
*521Instead of deciding the case on the grounds described above, the Court relies on prior decisions by the Court of Appeals to support an assumption that “without so stating, the court was exercising its supervisory powers to discipline the prosecutors of its jurisdiction.” Ante, at 505. Based on this assumption, the Court poses its own question for review: “whether, on this record, in a purported exercise of supervisory powers, a reviewing court may ignore the harmless-error analysis of Chapman.” Ibid. This question is not presented by the case. As noted, the Court of Appeals did not state that it was relying on its supervisory powers to reverse the convictions. It is sheer speculation for the Court to suggest that it was. Moreover, it is wholly inappropriate to address an important question concerning the scope of a federal appellate court’s supervisory powers based on the Court of Appeals’ decision in this case. Given the fact that the Court of Appeals did not expressly invoke its supervisory powers, it obviously also failed to detail the considerations that supported the exercise of such powers. The Court, therefore, has no explanation on which to base an analysis of the propriety of the Court of Appeals’ assumed exercise of its supervisory powers. The respondents’ effort to justify the Court of Appeals’ disposition of this case based on an exercise of the court’s supervisory powers provides no commission to this Court to decide important questions that are unnecessary to a decision in the case, are not presented by it, and cannot be analyzed carefully, if at all, based on the decision involved.
The problems posed by the Court of Appeals’ failure to explain its decision are evident in the Court’s discussion of supervisory powers. The Court suggests, for example, that “in this context, the integrity of the process carries less weight, for it is the essence of the harmless-error doctrine that a judgment may stand only when there is no ‘reasonable possibility that the [practice] complained of might have contributed to the conviction.’” Ante, at 506 (citation omitted). *522Unfortunately, we cannot be sure of the precise “context” in which this case arose. If, for example, the violation in this case was another in a long line of intentional violations of defendants’ rights by Government prosecutors, the “context” might be considerably different. An assessment of the weight carried by the “integrity of the process” also might be affected substantially by evidence of this sort. It is difficult to imagine that a series of intentional violations of defendants’ constitutional rights by Government prosecutors who are officers of the court charged with upholding the law would not have a considerable detrimental effect on the integrity of the process and call for judicial action designed to restore order and integrity to the process.
The Court also states that “deterrence is an inappropriate basis for reversal where, as here, the prosecutor’s remark is at most an attenuated violation of Griffin and where means more narrowly tailored to deter objectionable prosecutorial conduct are available.” Ibid, (footnotes omitted). Without disputing that a court of appeals generally should use means more narrowly tailored than reversal to deter improper pros-ecutorial conduct, there may be reasons why a court of appeals would reject the use of such means. Prior experience, for example, might have demonstrated the futility of relying on Department of Justice disciplinary proceedings.
The Court also states that “reversals of convictions under the court’s supervisory power must be approached ‘with some caution’ . . . and with a view toward balancing the interests involved . . . .” Ante, at 506-507. The Court goes on to state that the “Court of Appeals failed in this case to give appropriate — if, indeed, any — weight to these relevant interests.” Ante, at 507. According to the Court, the Court of Appeals “did not consider the trauma the victims of these particularly heinous crimes would experience in a new trial, forcing them to relive harrowing experiences now long past, or the practical problems of retrying these sensitive issues *523more than four years after the events.” Ibid. In the Court’s view, “[t]he conclusion is inescapable that the Court of Appeals focused exclusively on its concern that the prosecutors within its jurisdiction were indifferent to the frequent admonitions of the court.” Ibid. In my view, what the Court of Appeals did or did not do is a matter of sheer speculation. In the absence of an explanation, the Court has no way of knowing what considerations motivated the Court of Appeals. The speculative and unwarranted nature of the Court’s analysis is exacerbated by the fact that the Court must assume at the outset that the Court of Appeals in fact was relying on its supervisory powers.
The only thing of which we can be sure is that the Court of Appeals refused, without an adequate explanation, to apply the harmless-error doctrine. This error calls for vacating the judgment and remanding the case. See supra, at 520, and n. 2. It does not call for an extended discussion of the scope of an appellate court’s supervisory powers, an examination of the relationship between those powers and the harmless-error rule, a rejection of the exercise of those powers in the absence of an explanation to inform the analysis, or an application of the harmless-error rule by this Court in the first instance.
Although the Court’s opinion is not clear, it is possible that it could be read to establish a per se rule against use of the supervisory powers to reverse a conviction based on a harmless error. Compare ante, at 506, 509-510, n. 7, with ante, at 506-507, 509. See also ante, at 505 (“We hold that the harmless-error rule of Chapman . . . may not be avoided by an assertion of supervisory power, simply to justify a reversal of these criminal convictions” (emphasis supplied)). If the Court is attempting to establish a per se rule against using supervisory powers to reverse a conviction based on harmless error, the absence of an explanation by the Court of Appeals is not as great an impediment to its decision. The fact remains, however, that the question the Court chooses *524to resolve is not presented by the case and should not be reached. Although I would not reach the question, I do not believe that Chapman, or the fact that an error is harmless, necessarily precludes a court of appeals from exercising its supervisory powers to reverse a conviction.
In Chapman the Court addressed the question of whether a violation of the rule of Griffin v. California, 380 U. S. 609 (1965), can be held to be harmless. 386 U. S., at 20. In considering this question, the Court rejected a rule of automatic reversal. Id., at 22. We noted the prevalence of harmless-error statutes or rules and stated that these rules “serve a very useful purpose insofar as they block setting aside convictions for small errors or defects that have little, if any, likelihood of having changed the result of the trial.” Ibid. In this light, we concluded that “there may be some constitutional errors which in the setting of a particular case are so unimportant and insignificant that they may, consistent with the Federal Constitution, be deemed harmless, not requiring the automatic reversal of the conviction.” Ibid.3
In Connecticut v. Johnson, 460 U. S. 73 (1983), the plurality stated that “Chapman continued a trend away from the practice of appellate courts in this country and in England of ‘reversing] judgments for the most trivial errors.’” Id., at 82 (citation omitted). As the Court notes, the goal of the harmless-error rule is “ ‘to conserve judicial resources by enabling appellate courts to cleanse the judicial process of prejudicial error without becoming mired in harmless error.’” Ante, at 509 (citation omitted). Chapman also stands for the proposition that a criminal defendant is not entitled to reversal of his conviction if the constitutional violation at issue is subject to harmless-error analysis and, after the issue has been raised and the Government has carried its burden, the *525error is determined to be harmless within the meaning of Chapman. In this regard, the rule limits the remedies available to a criminal defendant whose rights have been violated, but it also advances the important social interest in not allowing harmless errors to upset otherwise valid criminal convictions.
The harmless-error rule announced in Chapman is based on important jurisprudential and social policies and generally should be applied to constitutional errors which it covers. This is not to suggest, however, that application of the harmless-error rule is a constitutional imperative; nothing in Chapman suggests that the rule always must be applied, or that convictions tainted only by harmless error never may be reversed. Chapman stands only for the proposition that certain constitutional guarantees do not themselves require reversal for harmless violations. If there is other authority, aside from the constitutional provisions violated in the case, that supports either a decision not to apply the rule or to reverse a conviction even though the error at issue is harmless, Chapman does not stand as a bar to such action. Federal statutes and state law are two such sources of authority.4 In my view, the supervisory powers of federal appellate courts provide another possible source of authority, under some carefully confined circumstances, either to forgo a harmless-error inquiry or to reverse a conviction even though the error at issue is harmless.
In McNabb v. United States, 318 U. S. 332 (1943), the Court stated that “[j]udicial supervision of the administration of criminal justice in the federal courts implies the duty of *526establishing and maintaining civilized standards of procedure and evidence.” Id., at 340. See also Thiel v. Southern Pacific Co., 328 U. S. 217, 225 (1946). In Cupp v. Naughten, 414 U. S. 141 (1973), the Court suggested that within the federal court system an “appellate court will, of course, require the trial court to conform to constitutional mandates, but it may likewise require it to follow procedures deemed desirable from the viewpoint of sound judicial practice although in nowise commanded by statute or by the Constitution.” Id., at 146. In Mesarosh v. United States, 352 U. S. 1 (1956), the Court observed: “This is a federal criminal case, and this Court has supervisory jurisdiction over the proceedings of the federal courts. If it has any duty to perform in this regard, it is to see that the waters of justice are not polluted.” Id., at 14. See also Communist Party v. Subversive Activities Control Board, 351 U. S. 115, 124 (1956) (“The untainted administration of justice is certainly one of the most cherished aspects of our institutions”). Other cases have acknowledged the duty of reviewing courts to preserve the integrity of the judicial process. In Donnelly v. DeChristoforo, 416 U. S. 637 (1974), the Court stated: “We do not, by this decision, in any way condone prosecutorial misconduct, and we believe that trial courts, by admonition and instruction, and appellate courts, by proper exercise of their supervisory power, will continue to discourage it.” Id., at 648, n. 23. Finally, in United States v. Payner, 447 U. S. 727 (1980), the Court noted that “the supervisory power serves the ‘twofold’ purpose of deterring illegality and protecting judicial integrity.” Id., at 736, n. 8.5
*527These cases indicate that the policy considerations supporting the harmless-error rule and those supporting the existence of an appellate court’s supervisory powers are not in irreconcilable conflict. Both the harmless-error rule and the exercise of supervisory powers advance the important judicial and public interest in the orderly and efficient administration of justice. Exercise of the supervisory powers also can further the strong public interest in the integrity of the judicial process. If Government prosecutors have engaged in a pattern and practice of intentionally violating defendants’ constitutional rights, a court of appeals certainly might be justified in reversing a conviction, even if the error at issue is harmless, in an effort to deter future violations. If effective as a deterrent, the reversal could avert further damage to judicial integrity. Admittedly, using the supervisory powers to reverse a conviction under these circumstances appears to conflict with the public’s interest in upholding otherwise valid convictions that are tainted only by harmless error. But it is certainly arguable that the public’s interests in preserving judicial integrity and in insuring that Government prosecutors, as its agents, refrain from intentionally violating defendants’ rights are stronger than its interest in upholding the conviction of a particular criminal defendant. Convictions are important, but they should not be protected at any cost.6
I have no occasion now to define the precise contours of supervisory powers or. to explore the circumstances in which *528using them to reverse a conviction based on harmless error might be appropriate. This much, however, is clear: A court of appeals should exercise its supervisory powers to reverse a conviction based on harmless error only in the most extreme circumstances and only after careful consideration, and balancing, of all the relevant interests.7 The policies supporting the harmless-error rule announced in Chapman should be given considerable, but not controlling, weight in that balance. In my view, there is nothing in Chapman that requires us to adopt a per se rule against using the supervisory powers to reverse a conviction based on harmless error. In light of the importance of the interests potentially at stake, it would be surprising if there were.8
It is possible that a court of appeals might not always have to protide a detailed explanation of a decision to invoke its supervisory powers. If, for example, the court in a prior case had announced a new rule adopted pursuant to its supervisory powers it may not have to explain again in a subsequent case the considerations that supported adoption of the rule. At the least, however, the court should invoke expressly the previously announced rule in order to make clear the basis for its decision.
As the Court points out, ante, at 504-505, the Court of Appeals discussed the continuing problem of improper prosecutorial comment in United States v. Rodriguez, 627 F. 2d 110 (CA7 1980), which is cited in the court’s opinion in this case. See 660 F. 2d, at 308. The Court states that the “Rodriguez court discussed, without explicitly adopting, the rule announced by the First Circuit in United States v. Flannery, 451 F. 2d 880, 882 (1971), that any prosecutorial reference to a defendant’s failure to testify is per se grounds for reversal unless the judge immediately instructs the jury that the defendant had a constitutional right not to testify and advises the jury that the prosecutor’s conduct was improper.” Ante, at 504-505. In fact, the Court of Appeals expressly declined “to adopt so strict a rule.” 627 F. 2d, at 113.
As Justice Powell noted in Connecticut v. Johnson, 460 U. S. 73 (1983), the question of whether an error is harmless is “[n]ormally ... a question more appropriately left to the courts below.” Id., at 102 (dissenting opinion). Accord, Moore v. Illinois, 434 U. S. 220, 232 (1977); Coleman v. Alabama, 399 U. S. 1, 11 (1970); Foster v. California, 394 U. S. 440, 444 (1969); United States v. Wade, 388 U. S. 218, 242 (1967). For reasons that are not clear, the Court declines to follow this practice in this case. See ante, at 510.
The Court noted that “there are some constitutional rights so basic to a fair trial that their infraction can never be treated as harmless error . . . .” 386 U. S.. at 23. See id., at 23. n. 8.
See Connecticut v. Johnson, 460 U. S., at 88 (Stevens, J., concurring in judgment) (Chapman “does not require a state appellate court to make a harmless-error determination; it merely permits the state court to do so in appropriate cases” (emphasis in original) (footnote omitted)). Similarly, Congress presumably could enact, consistent with the Constitution, a statute covering Griffin violations that would alter the rule in Chapman.
It is noteworthy that a majority of the Court in Hampton v. United States, 425 U. S. 484 (1976), a case involving the entrapment defense, suggested that supervisory powers possibly could be employed to bar conviction of a defendant based on outrageous police conduct even though the defendant might have been “predisposed.” See id., at 491, 493-495 (Powell, J., concurring in judgment); id., at 495, 497 (Brennan, J., dissenting).
The case is made even stronger if we consider, as the discussion in text does not, the interests of criminal defendants in having their constitutional rights protected. Whether or not an error ultimately is determined to be harmless, a defendant’s rights still have been violated. Criminal defendants have an even stronger interest in being protected from intentional violations of their constitutional rights, especially in view of the difficulties surrounding harmless-error inquiries. As the court noted in United States v. Rodriguez, 627 F. 2d, at 113, “[a] defendant’s liberty should not so often depend upon our struggle with the particular circumstances of a case to determine from a cold record whether or not the prosecutor’s remarks were harmless.”
Although the interests of a victim in a particular case are not relevant to determining whether to enforce the established rights of a criminal defendant, see Morris v. Slappy, ante, at 28-29, n. 10 (Brennan, J., concurring in result), the interests of a victim may be relevant to determining whether to invoke the supervisory powers to reverse a conviction in a particular ease even though the error is harmless. Whether a continuing problem calls for the exercise of supervisory powers is a different question from whether a particular case is an appropriate context in which to exercise those powers.
Like the Court, see ante, at 509-510, n. 7,1 do not reach the question of whether 28 U. S. C. § 2111 is coextensive with Chapman. In any event, I do not think that it necessarily forecloses the exercise of supervisory powers.