Reviewing courts normally disregard trial errors that are harmless. This case asks us to decide whether a federal habeas court should consider a trial error harmless when the court (1) reviews a state-court judgment from a criminal trial, (2) finds a constitutional error, and (3) is in grave doubt about whether or not that error is harmless. We recognize that this last mentioned circumstance, "grave doubt," is unusual. Normally a record review will permit a judge to make up his or her mind about the matter. And indeed a judge has an obligation to do so. But, we consider here the legal rule that governs the special circumstance in which record review leaves the conscientious judge in grave doubt about the likely effect of an error on the jury's verdict. (By "grave doubt" we mean that, in the judge's mind, the matter is so evenly balanced that he feels himself in virtual equipoise as to the harmlessness of the error.) We conclude that the uncertain judge should treat the error, not as if it were harmless, but as if it affected the verdict (i.e., as if it had a "substantial and injurious effect or influence in determining the jury's verdict").

* Robert O'Neal filed a federal habeas corpus petition challenging his state-court convictions for murder and other crimes. The Federal District Court agreed with several of his claims of constitutional trial error. On appeal the Sixth Circuit disagreed with the District Court, with one important exception. That exception focused on possible jury "confusion" arising out of a trial court instruction about the state of mind necessary for conviction combined with a related statement by a prosecutor. The Sixth Circuit assumed, for argument's sake, that the instruction (taken together with the prosecutor's statement) had indeed violated the Federal Constitution by misleading the jury. Nonetheless, the court disregarded the error on the ground that it was "harmless." 3 F.3d 143, 147 (1993).

The court's opinion sets forth the standard normally applied by a federal habeas court in deciding whether or not this kind of constitutional "trial" error is harmless, namely, whether the error " ' "had substantial and injurious effect or influence in determining the jury's verdict." ' " Id., at 145 (quoting Brecht v. Abrahamson, 507 U.S. ----, ----, 113 S.Ct. 1710, 1712, 123 L.Ed.2d 353 (1993) (quoting, and adopting, standard set forth in Kotteakos v. United States, 328 U.S. 750, 776, 66 S.Ct. 1239, 1253, 90 L.Ed. 1557 (1946))). But, rather than ask directly whether the record's facts satisfied this standard, the court seemed to refer to a burden of proof. Its opinion says that the habeas petitioner must bear the "burden of establishing" whether the error was prejudicial. 3 F.3d, at 145. As a practical matter, this statement apparently means that, if a judge is in grave doubt about the effect on the jury of this kind of error, the petitioner must lose. Thus, O'Neal might have lost in the Court of Appeals, not because the judges concluded that the error was harmless, but because the record of the trial left them in grave doubt about the effect of the error.

This Court granted certiorari to decide what the law requires in such circumstances. We repeat our conclusion: When a federal judge in a habeas proceeding is in grave doubt about whether a trial error of federal law had "substantial and injurious effect or influence in determining the jury's verdict," that error is not harmless. And, the petitioner must win.

As an initial matter, we note that we deliberately phrase the issue in this case in terms of a judge's grave doubt, instead of in terms of "burden of proof." The case before us does not involve a judge who shifts a "burden" to help control the presentation of evidence at a trial, but rather involves judges who apply a legal standard (harmlessness) to a record that the presentation of evidence is no longer likely to affect. In such a case, we think it conceptually clearer for the judge to ask directly, "Do I, the judge, think that the error substantially influenced the jury's decision?" than for the judge to try to put the same question in terms of proof burdens (e.g., "Do I believe the party has borne its burden of showing . . .?"). As Chief Justice Traynor said:

"Whether or not counsel are helpful, it is still the responsibility of the . . . court, once it concludes there was error, to determine whether the error affected the judgment. It must do so without benefit of such aids as presumptions or allocated burdens of proof that expedite fact-finding at the trial." R. Traynor, The Riddle of Harmless Error 26 (1970) (hereinafter Traynor).

The case may sometimes arise, however, where the record is so evenly balanced that a conscientious judge is in grave doubt as to the harmlessness of an error. See id., at 22-23. This is the narrow circumstance we address here.

Our legal conclusion—that in cases of grave doubt as to harmlessness the petitioner must win—rests upon three considerations. First, precedent supports our conclusion. As this Court has stated, "the original common-law harmless-error rule put the burden on the beneficiary of the error [here, the State] . . . to prove that there was no injury. . . ." Chapman v. California, 386 U.S. 18, 24, 87 S.Ct. 824, 828, 17 L.Ed.2d 705 (1967) (citing 1 J. Wigmore, Evidence § 21 (3d ed. 1940)). When this Court considered the doubt-as-to-harmlessness question in the context of direct review of a nonconstitutional trial error, it applied the same rule. In Kotteakos v. United States, the Court wrote:

"If, when all is said and done, the [court's] conviction is sure that the error did not influence the jury, or had but very slight effect, the verdict and the judgment should stand. . . . But if one cannot say, with fair assurance, after pondering all that happened without stripping the erroneous action from the whole, that the judgment was not substantially swayed by the error, it is impossible to conclude that substantial rights were not affected. The inquiry cannot be merely whether there was enough to support the result, apart from the phase affected by the error. It is rather, even so, whether the error itself had substantial influence. If so, or if one is left in grave doubt, the conviction cannot stand." 328 U.S., at 764-765, 66 S.Ct., at 1248 (emphasis added).

Id., at 776, 66 S.Ct., at 1248-49 (holding that error is not harmless if it had "substantial and injurious effect or influence" upon the jury). That is to say, if a judge has "grave doubt" about whether an error affected a jury in this way, the judge must treat the error as if it did so. See also United States v. Olano, 507 U.S. ----, ----, 113 S.Ct. 1770, ----, 123 L.Ed.2d 508 (1993) (stating that under Federal Rule of Criminal Procedure 52(a) the Government bears the "burden of showing the absence of prejudice"); United States v. Lane, 474 U.S. 438, 449, 106 S.Ct. 725, 732, 88 L.Ed.2d 814 (1986) (quoting Kotteakos as providing the proper harmless-error standard in cases of misjoinder and quoting " 'grave doubt' " language).

When this Court considered the same question in the context of direct review of a constitutional trial error, it applied the same rule. See Chapman, 386 U.S., at 24, 87 S.Ct., at 828 (holding that error is harmless only if "harmless beyond a reasonable doubt"). Indeed, the Chapman Court wrote that "constitutional error . . . casts on someone other than the person prejudiced by it a burden to show that it was harmless." Ibid.

We must concede that in Brecht v. Abrahamson this Court, in the course of holding that the more lenient Kotteakos harmless-error standard, rather than the stricter Chapman standard, normally governs cases of habeas review of constitutional trial errors, stated that habeas petitioners "are not entitled to habeas relief based on trial error unless they can establish that it resulted in 'actual prejudice.' " Brecht, 507 U.S., at ----, 113 S.Ct., at 1721 (emphasis added). This language, however, is not determinative. The issue in Brecht involved a choice of substantive harmless-error standards: the stricter Chapman, or the less strict Kotteakos, measure of harmlessness. Both of those cases had resolved the issue now before us the same way, placing the risk of doubt on the State. Moreover, the sentence from Brecht quoted above appears in a paragraph that adopts the very Kotteakos standard that we now apply. That paragraph does not explain why the Court would make an exception to the "grave doubt" portion of the Kotteakos standard. Furthermore, the Brecht opinion, in this respect, did not speak for a Court majority. Four Members of the Court, supporting application of Chapman's standard, dissented. And, Justice STEVENS, while a Member of the majority, stated explicitly that the Kotteakos standard applied in its entirety. 507 U.S., at ----, 113 S.Ct., at 1723 (STEVENS, J., concurring) (agreeing, in part, because that standard "places the burden on prosecutors to explain why those errors were harmless").

We further acknowledge that this Court, in Palmer v. Hoffman, 318 U.S. 109, 63 S.Ct. 477, 87 L.Ed. 645 (1943), said:

"He who seeks to have a judgment set aside because of an erroneous ruling carries the burden of showing that prejudice resulted." Id., at 116, 63 S.Ct., at 482.

But this pre-Kotteakos language, in context, referred to what the preceding sentence in Palmer described as "[m]ere 'technical errors.' " 318 U.S., at 116, 63 S.Ct., at 482. Kotteakos itself makes clear that "technical errors" may be different. It quotes an excerpt from the House Report accompanying the harmless-error statute (then 28 U.S.C. § 391) that says the statute casts

" 'upon the party seeking a new trial the burden of showing that any technical errors that he may complain of have affected his substantial rights.' " Kotteakos, 328 U.S., at 760, 66 S.Ct., at 1246 (quoting H.R.Rep. No. 913, 65th Cong., 3d Sess., 1 (1919)).

The Report, however, (as Kotteakos immediately points out) goes on to say that if the error is not "technical," if, instead,

" 'the error is of such a character that its natural effect is to prejudice a litigant's substantial rights, [then] the burden of sustaining a verdict will, notwithstanding this legislation rest upon the one who claims under it.' " Ibid.

The "grave doubt" language of Kotteakos itself makes clear that important trial errors, including any "constitutional violation, . . . fall in" this last mentioned category, and not "in the 'technical' category" that the House Report described. Brecht, supra, 507 U.S., at ----, 113 S.Ct., at 1723 (STEVENS, J., concurring); see Bruno v. United States, 308 U.S. 287, 294, 60 S.Ct. 198, 200-01, 84 L.Ed. 257 (1939) (Frankfurter, J.) (describing technical errors as those concerned with the "mere etiquette of trials and with the formalities and minutiae of procedure").

We also have examined the precedent upon which the State relies to support its view that appellants bear a "burden" of showing "prejudice" in civil cases. See, e.g., Erskine v. Consolidated Rail Corp., 814 F.2d 266 (CA6 1987); Flanigan v. Burlington Northern Inc., 632 F.2d 880 (CA8 1980); Creekmore v. Crossno, 259 F.2d 697 (CA10 1958). The State contends that, because a habeas proceeding, technically speaking, is a civil proceeding, see, e.g., Browder v. Director, Dept. of Corrections of Ill., 434 U.S. 257, 269, 98 S.Ct. 556, 563, 54 L.Ed.2d 521 (1978), this standard applies here. See Fed.Rule Civ.Proc. 61.

One problem with this argument lies in its failure to take into account the stakes involved in a habeas proceeding. Unlike the civil cases cited by the State, the errors being considered by a habeas court occurred in a criminal proceeding, and therefore, although habeas is a civil proceeding, someone's custody, rather than mere civil liability, is at stake. And, as we have explained, when reviewing errors from a criminal proceeding, this Court has consistently held that, if the harmlessness of the error is in grave doubt, relief must be granted. We hold the same here.

Moreover, precedent suggests that civil and criminal harmless-error standards do not differ in their treatment of grave doubt as to the harmlessness of errors affecting substantial rights. In Kotteakos, the Court interpreted the then-existing harmless-error statute, 28 U.S.C. § 391, now codified with minor change at 28 U.S.C. § 2111. See Kotteakos, 328 U.S., at 759, 66 S.Ct., at 1245 (explaining that the statute "grew out of widespread and deep conviction" that appellate courts had become " 'impregnable citadels of technicality' ") (citation omitted). That statute, by its terms, applied to both civil and criminal cases, and Kotteakos made no distinction, at least with respect to the question at issue here, between the two types of cases. See id., at 757, n. 9, 66 S.Ct., at 1244, n. 9; 11 C. Wright & A. Miller, Federal Practice and Procedure § 2883, p. 276 (1973) (hereinafter Wright & Miller). Similarly, the current harmless-error statute "traces its lineage" to § 391, and applies in both civil and criminal proceedings. See McDonough Power Equipment, Inc. v. Greenwood, 464 U.S. 548, 554, n. 4, 104 S.Ct. 845, 849 n. 4, 78 L.Ed.2d 663 (1984). And, more important for present purposes, the current harmless-error sections of the Federal Rules of Civil Procedure and the Federal Rules of Criminal Procedure (which use nearly identical language) both refer to § 391 as their statutory source. Compare Fed.Rule Crim.Proc. 52(a) (providing that "[a]ny error . . . which does not affect substantial rights shall be disregarded") with Fed.Rule Civ.Proc. 61 (providing that the court "must disregard any error . . . which does not affect the substantial rights of the parties"); see Advisory Committee's Notes on Fed.Rule Crim.Proc. 52(a), 18 U.S.C.App., p. 833 (referring to former § 391 as a source); Advisory Committee's Notes on Fed.Rule Civ.Proc. 61, 28 U.S.C.App., p. 676 (same). In fact, in recent cases, we have interpreted the nearly identical language of Rule 52(a) as treating instances of grave doubt just as we treat them here. See Olano, 507 U.S., at ----, ---- - ----, 113 S.Ct., at ----, ---- - ----; Lane, 474 U.S., at 449, 106 S.Ct., at 732 (quoting Kotteakos, 328 U.S., at 765, 66 S.Ct., at 1248). For these reasons, even if, for argument's sake, we were to assume that the civil standard for judging harmlessness applies to habeas proceedings (despite the fact that they review errors in state criminal trials), it would make no difference with respect to the matter before us. For relevant authority rather clearly indicates that, either way, the courts should treat similarly the matter of "grave doubt" regarding the harmlessness of errors affecting substantial rights, and as Kotteakos provides.

Second, our conclusion is consistent with the basic purposes underlying the writ of habeas corpus. As we have said, we are dealing here with an error of constitutional dimension—the sort that risks an unreliable trial outcome and the consequent conviction of an innocent person. See Brecht, 507 U.S., at ----, 113 S.Ct. at ---- - ---- (O'CONNOR, J., dissenting). We also are assuming that the judge's conscientious answer to the question, "But, did that error have a 'substantial and injurious effect or influence' on the jury's decision?" is, "It is extremely difficult to say." In such circumstances, a legal rule requiring issuance of the writ will, at least often, avoid a grievous wrong—holding a person "in custody in violation of the Constitution . . . of the United States." 28 U.S.C. §§ 2241(c)(3), 2254(a). Such a rule thereby both protects individuals from unconstitutional convictions and helps to guarantee the integrity of the criminal process by assuring that trials are fundamentally fair. See Traynor 23 ("In the long run there would be a closer guard against error at the trial, if . . . courts were alert to reverse, in case of doubt, for error that could have contaminated the judgment"). By way of contrast, the opposite rule, denying the writ in cases of grave uncertainty, would virtually guarantee that many, in fact, will be held in unlawful custody—contrary to the writ's most basic traditions and purposes. And, it would tell judges who believe individuals are quite possibly being held "in custody in violation of the Constitution" that they cannot grant relief.

We concede that this opposite rule (denying the writ) would help protect the State's interest in the finality of its judgments and would promote federal-state comity. It would avoid retrials, some of which, held so late in the day, may lead to freedom for some petitioners whose initial convictions were in fact unaffected by the errors that took place at their initial trials. The State's interest in avoiding retrial of this latter category of individuals is legitimate and important. But this interest is somewhat diminished by the legal circumstance that the State normally bears responsibility for the error that infected the initial trial. And, if one assumes (1) that in cases of grave doubt, the error is at least as likely to have been harmful in fact as not, and (2) that retrial will often (or even sometimes) lead to reconviction, then that state interest is further diminished by a factual circumstance: the number of acquittals wrongly caused by grant of the writ and delayed retrial (the most serious harm affecting the State's legitimate interests) will be small when compared with the number of persons whom this opposite rule (denying the writ) would wrongly imprison or execute. On balance, we must doubt that the law of habeas corpus would hold many people in prison "in violation of the Constitution," for fear that otherwise a smaller number, not so held, may eventually go free.

Third, our rule has certain administrative virtues. It is consistent with the way that courts have long treated important trial errors. See, e.g., Olano, supra; Lane, supra; Chapman, 386 U.S., at 24, 87 S.Ct. at 828; Kotteakos, supra; see also 11 Wright & Miller § 2883. In a highly technical area such as this one, consistency brings with it simplicity, a body of existing case law available for consultation, see Brecht, supra, at ----, 113 S.Ct., at ---- - ----, and a consequently diminished risk of further, error-produced, proceedings. Moreover, our rule avoids the need for judges to read lengthy records to determine prejudice in every habeas case. These factors are not determinative, but offer a practical caution against a legal rule that, in respect to precedent and purpose, would run against the judicial grain.

The State makes one additional argument. It points to language in the habeas corpus statute that says the federal courts "shall entertain an application for a writ of habeas corpus . . . only on the ground that [a petitioner] is in custody in violation of the Constitution or laws or treaties of the United States." 28 U.S.C. § 2254(a).

See § 2241(c)(3). If a "violation of the Constitution" is harmless, the State adds, then there is no causal connection between "violation" and "custody," and the prisoner is not "in custody in violation of the Constitution." And, by analogy to tort law, the State contends that, because the habeas petitioner is in the position of plaintiff, he must prove this causal connection. Whatever force there may be to this argument is countered by the equally persuasive analogy to affirmative defenses, on which the party in the position of defendant (here the State) bears the risk of equipoise. And, to read the statute as the State suggests would run counter to the principle of Kotteakos that when an error's natural effect is to prejudice substantial rights and the court is in grave doubt about the harmlessness of that error, the error must be treated as if it had a "substantial and injurious effect" on the verdict. See Kotteakos, 328 U.S., at 764-765, 776, 66 S.Ct., at 1247-48, 1253.

We do not see what in the language of the statute tells a court that it should treat a violation as harmless when it is in grave doubt about its harmlessness. One might as easily infer the opposite—that the statute leaves the matter of harmlessness as a kind of affirmative defense—from the absence, in the Habeas Corpus Rules' form petition, of any space for a "lack of harmlessness" allegation. See 28 U.S.C. § 2254 Rule 2(c) (providing in part that a habeas petition "shall be in substantially the form annexed to these rules"). Or, one might as easily infer neutrality on the point from the statute's command that the court dispose of the petition "as law and justice require." 28 U.S.C. § 2243. Ultimately, we find no significant support for either side in any of this language. When faced with such gaps in the habeas statute, we have "look[ed] first to the considerations underlying our habeas jurisprudence, and then determine[d] whether the proposed rule would advance or inhibit these considerations by weighing the marginal costs and benefits of its application on collateral review." Brecht, 507 U.S., at ----, 113 S.Ct., at 1719. We have done that in this case, and for the reasons set forth above, see supra, at ___-___, we conclude that, when a habeas court is in grave doubt as to the harmlessness of an error that affects substantial rights, it should grant relief.

For these reasons, the judgment of the Court of Appeals is vacated, and the case is remanded for further proceedings consistent with this opinion.

It is so ordered.

Justice THOMAS, with whom Chief Justice REHNQUIST and Justice SCALIA join, dissenting.

In my view, a federal habeas court may not upset the results of a criminal trial unless it concludes both that the trial was marred by a violation of the Constitution or a federal statute and that this error was harmful. Because the Court concludes otherwise, I respectfully dissent.

* Though the majority begins with an examination of precedent construing the federal harmless-error statute, 28 U.S.C. § 2111, the proper place to begin is with the statute governing habeas relief for prisoners in state custody. After all, the petitioner does not seek relief under the harmless-error statute.

Where a state prisoner is concerned, a writ of habeas corpus may issue only when that prisoner is "in custody in violation of the Constitution or laws or treaties of the United States." 28 U.S.C. §§ 2241(c)(3), 2254(a). It is not enough that the habeas petitioner is in custody and that some violation of the Constitution or a federal statute occurred at trial; as amicus curiae the Solicitor General correctly argues, the statute requires a causal link between the violation and the custody. Quite obviously, a habeas petitioner who proves that a trivial ("harmless") error occurred at trial will not secure habeas relief because such an error could not be said to have been a cause of the custody. Notwithstanding the error, the petitioner would have been in custody and thus relief is unwarranted. Even the majority implicitly agrees that causation is necessary, for otherwise it would have no need to discuss harmful errors as opposed to mere errors.

The habeas petitioner comes to federal court as a plaintiff. Because the plaintiff "seeks to change the present state of affairs," he "naturally should be expected to bear the risk of failure of proof or persuasion." 2 McCormick on Evidence § 337, at 428 (J. Strong, 4th ed. 1992). Part of that burden is the requirement that the plaintiff show that the defendant's actions caused harm. In other areas of the law, the plaintiff almost invariably bears the burden of persuasion with respect to whether the defendant's actions caused harm. See, e.g., Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 669, 109 S.Ct. 2115, 2131, 104 L.Ed.2d 733 (1989) (STEVENS, J., dissenting) ("In the ordinary civil trial, the plaintiff bears the burden of persuading the trier of fact that the defendant has harmed her"); 2 Restatement (Second) of Torts § 433B(1) (1965) ("the burden of proof that the tortious conduct of the defendant has caused the harm to the plaintiff is upon the plaintiff"). Establishing causation is thus an essential element of the plaintiff's case in chief. Under the majority's rationale, however, the habeas petitioner need not prove causation at all; once a prisoner establishes error, the government must affirmatively persuade the court of the harmlessness of that error. Ante, at __. Without explaining why it favors habeas plaintiffs over other plaintiffs, the Court thus treats the question of causation as an affirmative defense.

Requiring the habeas petitioner to bear the risk of non-persuasion not only accords with the usual rules of litigation, but also is compelled by what we have said about the nature of habeas relief. "When the process of direct review . . . comes to an end, a presumption of finality and legality attaches to the conviction and sentence." Barefoot v. Estelle, 463 U.S. 880, 887, 103 S.Ct. 3383, 3391-92, 77 L.Ed.2d 1090 (1983). We have acknowledged that the "writ strikes at finality," one of the "law's very objects," McCleskey v. Zant, 499 U.S. 467, 491, 111 S.Ct. 1454, 1468, 113 L.Ed.2d 517 (1989), and that when a habeas petitioner obtains a new trial, the government is put at a disadvantage "through the 'erosion of memory' and 'dispersion of witnesses' that occur with the passage of time," Kuhlmann v. Wilson, 477 U.S. 436, 453, 106 S.Ct. 2616, 2627, 91 L.Ed.2d 364 (1986) (plurality opinion) (quoting Engle v. Isaac, 456 U.S. 107, 127-128, 102 S.Ct. 1558, 1571-72, 71 L.Ed.2d 783 (1982)). Our habeas cases indicate that upsetting the finality of judgments should be countenanced only in rare instances. See, e.g., Brecht v. Abrahamson, 507 U.S. ----, ---- [113 S.Ct. 1710, 1719, 123 L.Ed.2d 353] (1993) (noting that "the writ of habeas corpus has historically been regarded as an extraordinary remedy").

We have ample cause to be wary of the writ. Our criminal law does not routinely punish the innocent. Instead, our Constitution requires proof of guilt beyond a reasonable doubt. See In Re Winship, 397 U.S. 358, 90 S.Ct. 1068, 25 L.Ed.2d 368 (1970). As a result, the overwhelming majority of the innocent will never reach the habeas stage, since they will not have been found guilty at trial. Appeals and possible state postconviction relief further reduce the possibility that an innocent is in custody. The presumption of finality that we apply in habeas proceedings is therefore well founded.

Our habeas jurisprudence has also been informed by a proper recognition of the affront to a state when federal courts conduct habeas review. Habeas review " 'disturbs the State's significant interest in repose for concluded litigation, denies society the right to punish some admitted offenders, and intrudes on state sovereignty to a degree matched by few exercises of federal judicial authority.' " Duckworth v. Eagan, 492 U.S. 195, 210, 109 S.Ct. 2875, 2884, 106 L.Ed.2d 166 (1989) (O'CONNOR, J., concurring) (quoting Harris v. Reed, 489 U.S. 255, 282, 109 S.Ct. 1038, 1054, 103 L.Ed.2d 308 (1989) (KENNEDY, J., dissenting)). See also McCleskey, 499 U.S., at 491, 111 S.Ct., at 1468-69; Engle, 456 U.S., at 128, 102 S.Ct., at 1572. Where the habeas court cannot say that an error resulted in harm, it seems particularly disrespectful to resolve doubts against the propriety of state court judgments.

Our "harmless-error" inquiry in the habeas context concerns whether an error " 'had a substantial and injurious effect or influence in determining the jury's verdict.' " Brecht, 507 U.S., at ---- [113 S.Ct. at 1714] (quoting Kotteakos v. United States, 328 U.S. 750, 776, 66 S.Ct. 1239, 1253, 90 L.Ed. 1557 (1946)). As alluded to earlier, supra, at ___, this test consists of a causation inquiry—was the error a cause of the conviction. Thus, like all plaintiffs, the habeas plaintiff must show causation if he is to succeed.

The Court derives its contrary rule from cases construing the harmless-error statute, the purposes underlying the writ of habeas corpus, and the virtue of administrative consistency that stems from following established precedent. The Court's analysis is unpersuasive.

The Court begins by examining harmless-error practice in the context of direct criminal appeals. I do not quarrel with the majority's conclusion that once an error has been shown on direct appeal, the government must demonstrate that it was harmless if the conviction is to stand. See ante at 4-5, (citing Kotteakos, supra, 328 U.S., at 764-765, 776, 66 S.Ct. at 1247-48, 1253; Chapman v. California, 386 U.S. 18, 24, 87 S.Ct. 824, 828, 17 L.Ed.2d 705 (1967); and United States v. Olano, 507 U.S. ----, ----, ---- [113 S.Ct. 1770, 1778, 1781, 123 L.Ed.2d 508] (1993)). Still, neither the harmless-error statute (which was the subject of Kotteakos ) nor the rules governing harmless errors in District Courts (discussed in Olano ) apply to habeas review of errors that occurred in a prior case. See 28 U.S.C. § 2111; Fed.R.Crim.Proc. 52(a); see also Fed.R.Civ.Proc. 61.

To be sure, we have borrowed the applicable standard for judging harmlessness in habeas from cases interpreting the federal harmless-error statute. See, e.g., Brecht, 507 U.S., at ---- [113 S.Ct. at 1718]. Applying harmless-error analysis makes sense, because a trivial error could not be said to cause custody and thus warrant habeas relief. But the harmless-error statute and rules do not apply of their own force in the habeas cases, and so the harmless-error precedents relied upon by the majority are certainly not dispositive. Indeed, Brecht itself—despite adopting the standard for harmlessness set out in Kotteakos—departed from Kotteakos by placing the burden upon the habeas petitioner to "establish" that this standard has been met. See 507 U.S., at ---- [113 S.Ct. at 1722].

If we are to look at cases examining the harmless-error statute, I would think that civil cases would be of greater relevance. As the Court admits, habeas is a civil proceeding. See ante, at __ (citing Browder v. Director, Dept. of Corrections of Illinois, 434 U.S. 257, 269, 98 S.Ct. 556, 563, 54 L.Ed.2d 521 (1978)). Although the Court acknowledges that Palmer v. Hoffman, 318 U.S. 109, 63 S.Ct. 477, 87 L.Ed. 645 (1943), put the burden on the party claiming prejudice to demonstrate it, the Court dismisses Palmer as a pre-Kotteakos case about technical errors. See ante, at __. But Kotteakos did not purport to overrule Palmer. Nor is it true that the rule in Palmer is limited to cases involving technical errors. Palmer merely quoted former 28 U.S.C. § 391 (the predecessor to the modern § 2111, and the statute at issue in Kotteakos as well), which itself referred to "technical errors." Palmer held that the party seeking relief from a judgment because of an erroneous ruling "carries the burden of showing that prejudice resulted"; it did not say that only those challenging "technically" erroneous rulings were so encumbered. See 318 U.S. at 116, 63 S.Ct. at 481-82. Accordingly, most of the Courts of Appeals that have considered the issue place the burden of showing prejudice on the civil appellant, just as Palmer did. See, e.g., Smith v. Wal-Mart Stores (No. 471), 891 F.2d 1177, 1180 (CA5 1990) (per curiam); United States v. Killough, 848 F.2d 1523, 1527 (CA11 1988); United States v. Seaboard Surety Co., 817 F.2d 956, 964 (CA2), cert. denied, 484 U.S. 855, 108 S.Ct. 161, 98 L.Ed.2d 115 (1987); see also ante, at __ (citing cases from the Sixth, Eighth and Tenth Circuits). But see Barth v. Gelb, 2 F.3d 1180, 1188 (CADC 1993) (quoting the "grave doubt" language of Kotteakos, 328 U.S. at 765, 66 S.Ct. at 1248).

The Court concludes that Palmer and these cases may be disregarded because the federal harmless-error statute, 28 U.S.C. § 2111, makes no distinction between civil and criminal cases; since the rule in the criminal context places the burden of persuasion on the government, the Court decides that the same should be true in the civil context. Ante, at __. But the majority's syllogism could just as easily be turned against the result it reaches. Authority in the civil context assigns the risk of non-persuasion to the party alleging error, and since the statute draws no distinction between civil and criminal cases, we might just as easily conclude that the civil rule should be followed in the criminal context. The Court's reasoning yields no determinate answer.

As indicated above, however, the harmless-error provisions do not actually apply in habeas cases anyway. We have no occasion to harmonize the harmless-error cases by overruling Palmer and by rejecting the practice that prevails in the majority of the Courts of Appeals that have considered the issue, as the Court does today.

The Court's second claim is that its "conclusion is consistent with the basic purposes underlying the writ of habeas corpus." Ante, at __. As part of its argument, the Court lays claim to the moral high ground: "We are dealing here with an error of constitutional dimension—the sort that risks an unreliable trial outcome and the consequent conviction of an innocent person." Ibid. The Court suggests that when there is a grave doubt about the harmfulness of an error, "a legal rule requiring issuance of the writ will, at least often, avoid a grievous wrong—holding a person 'in custody in violation of the Constitution . . . of the United States.' " Id., at ___ (quoting 28 U.S.C. §§ 2241(c)(3), 2254(a)).

The Court concedes that there are other interests at stake—a state's interest in the finality of its judgments and the promotion of federal-state comity, see ante, at __—but goes on to set these principles aside. The Court concludes that the state's interest in finality, while "legitimate and important", ibid., is diminished by the fact that "the number of acquittals wrongly caused by grant of the writ and delayed retrial . . . will be small when compared with the number of persons whom [the] opposite rule . . . would wrongly imprison or execute," ante, at __.

Despite its rhetoric, the Court itself is merely balancing the costs and benefits associated with disturbing judgments when a court is in grave doubt about harm. The Court decides that the possibility of unlawful custody should lead to the adoption of its grave doubt rule. But because the Court draws the line at "grave doubt" rather than "significant doubt" or "any doubt," it is not willing to go as far as it must in order to ensure that no one is unlawfully imprisoned. Thus, under the majority's assumptions, even its own rule will guarantee that "many, in fact, will be held in unlawful custody." Id., at 9.

It is important to recognize, moreover, that when the Court discusses erroneous imprisonments and executions, it is not addressing questions of innocence or guilt. The standard for judging harmlessness in habeas cases certainly does not turn on the innocence of the habeas petitioner. In fact, the Court's rule applies only when the habeas court cannot make up its mind about whether a jury would have entertained any reasonable doubt about the defendant's guilt. Though the majority seems to suggest otherwise, it certainly will not be true that in half of such cases, the state will have unjustly imprisoned an innocent person.

Citing Kotteakos, Chapman, and other cases, the Court concludes that its rule will be easier to administer because it is consistent with the way courts have treated grave doubts about harm. Ante, at __. As indicated above, Palmer and the majority view in the Courts of Appeal provide an equally attractive rule that is consistent with longstanding practice. As for the Court's assertion that its rule eliminates "the need for judges to read lengthy records to determine prejudice in every habeas case", ibid., I thought it settled that "it is the duty of a reviewing court to consider the trial record as a whole" when conducting a harmless-error analysis, United States v. Hasting, 461 U.S. 499, 509, 103 S.Ct. 1974, 1980, 76 L.Ed.2d 96 (1983). Surely a judge cannot, in the midst of reading a record, declare himself to be in grave doubt, stop reviewing, and issue the writ. Because further review may always disturb the judge's current view of the error, the judge cannot stop until he finishes reviewing the relevant portions of the record. Indeed, given that further review always has the potential to resolve any grave doubt, one is tempted to require a judge to continue to read and reread the relevant portions of the record until his grave doubts dissipate.

Fortunately, the rule announced today will affect only a minuscule fraction of cases. Even when there is a close question about whether an error was harmful, the conscientious judge ordinarily should make a ruling as to harm. The Court's rule is not a means for judges to escape difficult decisions; it applies only in that "special circumstance" in which a judge, after a thorough review of the record, remains in equipoise. See ante, at __,.

The rule has such limited application that it most likely will have no effect on this case. The majority suggests that O'Neal "might have lost in the Court of Appeals, not because the judges concluded that [any supposed] error was harmless, but because the record of the trial left them in grave doubt about the effect of the error." Id., at ___. The Sixth Circuit did observe that "[t]he habeas petitioner bears the burden of establishing . . . prejudice." O'Neal v. Morris, 3 F.3d 143, 145 (1993). But the Court of Appeals did not refer again to this burden and did not appear to rely on it in reaching a decision. See id., at 147. That we chose this case to establish a "grave doubt" rule is telling: cases in which habeas courts are in equipoise on the issue of harmlessness are astonishingly rare.

Though the question that the Court decides today will have very limited application, I believe that the Court gives the wrong answer to that question.

Accordingly, I respectfully dissent.