with whom Justice Brennan and Justice Marshall join, dissenting.
The rationale of our decision in Bruton v. United States, 391 U. S. 123, 135-136 (1968), applies without exception to all *212inadmissible confessions that are “powerfully incriminating.” Today, however, the Court draws a distinction of constitutional magnitude between those confessions that directly identify the defendant and those that rely for their inculpa-tory effect on the factual and legal relationships of their contents to other evidence before the jury. Even if the jury’s indirect inference of the defendant’s guilt based on an inadmissible confession is much more devastating to the defendant’s case than its inference from a direct reference in the co-defendant’s confession, the Court requires the exclusion of only the latter statement. This illogical result demeans the values protected by the Confrontation Clause. Moreover, neither reason nor experience supports the Court’s argument that a consistent application of the rationale of the Bruton case would impose unacceptable burdens on the administration of justice.
I
It is a “basic premise” of the Confrontation Clause that certain kinds of hearsay “are at once so damaging, so suspect, and yet so difficult to discount, that jurors cannot be trusted to give such evidence the minimal weight it logically deserves, whatever instructions the trial judge might give.”1 *213This constitutionally mandated skepticism undergirds the Bruton holding and is equally applicable to this case. The Court framed the issue in Bruton as “whether the conviction of a defendant at a joint trial should be set aside although the jury was instructed that a codefendant’s confession inculpating the defendant had to be disregarded in determining his guilt or innocence.” 391 U. S., at 123-124. We answered that question in the affirmative, noting that the Sixth Amendment is violated “where the powerfully incriminating extrajudicial statements of a codefendant, who stands accused side-by-side with the defendant, are deliberately spread before the jury in a joint trial.” Id., at 135-136.
Today the Court nevertheless draws a line between co-defendant confessions that expressly name the defendant and those that do not. The Court relies on the presumption that in the latter category “it is a less valid generalization that the jury will not likely obey the instruction to disregard the evidence.” Ante, at 208. I agree; but I do not read Bruton to require the exclusion of all codefendant confessions that do not mention the defendant.2 Some such confessions may not have any significant impact on the defendant’s case. But others will. If we presume, as we must, that jurors give their full and vigorous attention to every witness and each item of evidence, the very acts of listening and seeing will sometimes lead them down “the path of inference.” Indeed, the Court tacitly acknowledges this point; while the Court speculates that the judge’s instruction may dissuade the jury *214from making inferences at all, it also concedes the probability of their occurrence, arguing that there is no overwhelming probability that jurors will be unable to “disregard an incriminating inference.” Ibid. Bruton has always required trial judges to answer the question whether a particular confession is or is not “powerfully incriminating” on a case-by-case basis; they should follow the same analysis whether or not the defendant is actually named by his or her codefendant.
Instructing the jury that it was to consider Benjamin Williams’ confession only against him, and not against Clarissa Marsh, failed to guarantee the level of certainty required by the Confrontation Clause. The uncertainty arose because the prosecution’s case made it clear at the time Williams’ statement was introduced that the statement would prove “powerfully incriminating” of the respondent as well as of Williams himself. There can be absolutely no doubt that spreading Williams’ carefully edited confession before the jury intolerably interfered with the jury’s solemn duty to treat the statement as nothing more than meaningless sounds in its consideration of Marsh’s guilt or innocence.
At the time that Williams’ confession was introduced, the evidence already had established that respondent and two men committed an armed robbery in the course of which the two men killed two persons and shot a third. Ante, at 202. There was a sharp dispute, however, on the question whether respondent herself intended to commit a robbery in which murder was a foreseeable result, or knew that the two men planned to do so. The quantum of evidence admissible against respondent was just sufficient to establish this intent and hence to support her conviction. As the Court of Appeals explained:
“[T]he issue is whether the evidence was sufficient to show that Marsh aided and abetted the assault with the specific intent to murder Knighton or with the knowledge that Martin had this specific intent.... Marsh’s case presents a much closer question on this issue than *215does Williams’. There was no testimony indicating she harbored an intent to murder Knighton, nor was there any showing that she heard Martin’s statements regarding the need to ‘hurt’ or ‘take out’ the victims. There was, in addition, no testimony placing her in the basement, the scene of the shootings. The evidence does indicate, viewed in the light most favorable to the prosecution, that she was aware that Williams and Martin were armed, that she served as a guard or ‘lookout’ at the door, that she prevented an attempted escape by Knighton, and that she was given the paper bag thought to contain the proceeds of a robbery. The evidence also indicates that Marsh knew Scott, supporting the inference that it was Marsh who allowed Martin to gain entrance. While it is a close question, we believe the evidence presented at the time of the motion was sufficient to survive a motion for directed verdict.” 781 F. 2d 1201, 1204 (CA6 1986) (emphasis omitted).
In the edited statement that the jury was instructed not to consider against Marsh, Williams described the conversation he had with Kareem Martin while they were in a car driving to their victims’ residence. In that conversation, Martin stated that “he would have to take them out after the robbery.” See ante, at 203, n. 1. The State’s principal witness had testified that Martin and Marsh arrived at the victims’ house together. The jury was therefore certain to infer from the confession that respondent had been in the car and had overheard the statement by Martin. Viewed in the total context of the trial evidence, this confession was of critical importance because it was the only evidence directly linking respondent with the specific intent, expressed before the robbery, to kill the victims afterwards.3 If Williams had taken *216the witness stand and testified, respondent’s lawyer could have cross-examined him to challenge his credibility and to establish or suggest that the car radio was playing so loudly that Marsh could not have overheard the conversation between the two men from the backseat. An acknowledgment of the possibility of such facts by Williams would have done much more to eliminate the certainty beyond a reasonable doubt that Marsh knew about the murder plan than could possibly have been achieved by the later testimony of respondent herself. Moreover, the price respondent had to pay in order to attempt to rebut the obvious inference that she had overheard Martin was to remind the jury once again of what he had said and to give the prosecutor a further opportunity to point to this most damaging evidence on the close question of her specific intent. See ante, at 205, n. 2.
The facts in this case are, admittedly, different from those in Bruton because Williams’ statement did not directly mention respondent. Thus, instead of being “incriminating on its face,” ante, at 208, it became so only when considered in connection with the other evidence presented to the jury. The difference between the facts of Bruton and the facts of this case does not eliminate their common, substantial, and constitutionally unacceptable risk that the jury, when resolving *217a critical issue against respondent, may have relied on impermissible evidence.4
II
The facts that joint trials conserve prosecutorial resources, diminish inconvenience to witnesses, and avoid delays in the administration of criminal justice have been well known for a long time. See United States v. Lane, 474 U. S. 438, 449 (1986) (quoting Bruton, 391 U. S., at 134). It is equally well known that joint trials create special risks of prejudice to one of the defendants, and that such risks often make it necessary to grant severances. See Bruton, 391 U. S., at 131; Fed. Rule Crim. Proc. 14 (Relief from Prejudicial Joinder). The Government argues that the costs of requiring the prosecution to choose between severance and not offering the co-defendant’s confession at a joint trial outweigh the benefits to the defendant. Brief for United States as Amicus Curiae 22. On the scales of justice, however, considerations of fairness normally outweigh administrative concerns.
In the Bruton case the United States argued that the normal “benefits of joint proceedings should not have to be sacri*218ficed by requiring separate trials in order to use the confession against the declarant.” See 391 U. S., at 134. The Court endorsed the answer to this argument that Judge Lehman of the New York Court of Appeals had previously made in his dissenting opinion in People v. Fisher, 249 N. Y. 419, (432, 164 N. E. 336, 341 (1928):
“We still adhere to the rule that an accused is entitled to confrontation of the witnesses against him and the right to cross-examine them .... We destroy the age-old rule which in the past has been regarded as a fundamental principle of our jurisprudence by a legalistic formula, required of the judge, that the jury may not consider any admissions against any party who did not join in them. We secure greater speed, economy and convenience in the administration of the law at the price of fundamental principles of constitutional liberty. That price is too high.”
The concern about the cost of joint trials, even if valid, does not prevail over the interests of justice. Moreover, the Court’s effort to revive this concern in a state criminal case rests on the use of irrelevant statistics. The Court makes the startling discovery that joint trials account for “almost one-third of federal criminal trials in the past five years.” Ante, at 209. In the interest of greater precision, the Court might have stated that there were 10,904 federal criminal trials involving more than one defendant during that 5-year period.5 The Court might have added that the data base from which that figure was obtained does not contain any information at all to show the number of times that confessions were offered in evidence in those 10,904 federal cases.6 The *219relevance of this data is also difficult to discern because all of the cases in this Court that involved joint trials conducted after Bruton was decided, in which compliance with the rule of that case was at issue, appear to have originated in a state court. Federal prosecutors seem to have had little difficulty, in conducting the literally thousands of joint trials to which the Court points, in maintaining “both the efficiency and the fairness of the criminal justice system” that the Court speculates will occur if Bruton’s reasoning is applied to this case. See ante, at 210. Presumably the options of granting immunity, making plea bargains, or simply waiting until after a confessing defendant has been tried separately before trying to use his admissions against an accomplice have enabled the Federal Government to enforce the criminal law without sacrificing the basic premise of the Confrontation Clause.7
*220The Court also expresses concern that trial judges will be unable to determine whether a codefendant’s confession that does not directly mention the defendant and is inadmissible against him will create a substantial risk of unfair prejudice. In most such cases the trial judge can comply with the dictates of Bruton by postponing his or her decision on the admissibility of the confession until the prosecution rests, at which time its potentially inculpatory effect can be evaluated in the light of the government’s entire case. The Court expresses concern that such a rule would enable “manipulation by the defense,” see ante, at 209, by which the Court presumably means the defense might tailor its evidence to make sure that a confession which does not directly mention the defendant is deemed powerfully incriminating when viewed in light of the prosecution’s entire case. As a practical matter, I cannot believe that there are many defense lawyers who would deliberately pursue this high-risk strategy of “manipulating” their evidence in order to enhance the prejudicial impact of a codefendant’s confession. Moreover, a great many experienced and competent trial judges throughout the Nation are fully capable of managing cases and supervising counsel in order to avoid the problems that seem insurmountable to appellate judges who are sometimes distracted by illogical distinctions and irrelevant statistics.
I respectfully dissent.8
Bruton v. United States, 391 U. S. 123, 138 (1968) (Stewart, J., concurring) (emphasis in original). Judge Learned Hand and Justice Frankfurter also would recognize that the admission of Williams’ confession, even with limiting instructions, placed too great a strain upon the jury’s ability to exclude this evidence from its consideration of respondent’s innocence or guilt. As we noted in Bruton:
“Judge Hand addressed the subject several times. The limiting instruction, he said, is a ‘recommendation to the jury of a mental gymnastic which is beyond, not only their powers, but anybody’s else,’ Nash v. United States, 64 F. 2d 1006, 1007; ‘Nobody can indeed fail to doubt whether the caution is effective, or whether usually the practical result is not to let in hearsay,’ United States v. Gottfried, 165 F. 2d 360, 367; ‘it is indeed very hard to believe that a jury will, or for that matter can, in practice observe the admonition,’ Delli Paoli v. United States, 229 F. 2d 319, 321. Judge Hand referred to the instruction as a ‘placebo,’ medically defined as ‘a medicinal lie.’ ” 391 U. S., at 132, n. 8.
*213In a similar vein, Justice Frankfurter observed:
‘The Government should not have the windfall of having the jury be influenced by evidence against a defendant which, as a matter of-law, they should not consider but which they cannot put out of their minds.” Delli Paoli v. United States, 352 U. S. 232, 248 (1957) (dissenting opinion).
Indeed, I have no doubt that there are some codefendant confessions that expressly mention the defendant but nevertheless need not be excluded under Bruton because they are not prejudicial.
The Court assumes that the confession did not incriminate respondent at the time the confession was introduced. I disagree. Cynthia Knighton had already testified that respondent and Kareem Martin had arrived at *216the victims’ residence together, and that respondent admitted Williams to the house a few minutes later. In his statement Williams said:
‘We then drove over to this house and parked the car across the big street near the house. The plan was that I would wait in the car in front of the house and then I would move the car down across the big street because he didn’t want anybody to see the car. Okay, Kareem went up to the house and went inside. A couple of minutes later I moved the car and went up to the house.” Ante, at 203, n. 1.
It is unrealistic to believe that the jury would assume that respondent did not accompany the two men in the car but had just magically appeared at the front door of the apartment at the same time that Martin did.
It is worth noting that the dissenting opinion in Bruton did not regard the Court’s decision as limited to eodefendant confessions expressly implicating the defendant:
“I would suppose that it will be necessary to exclude all extrajudicial confessions unless all portions of them which implicate defendants other than the declarant are effectively deleted. Effective deletion will probably require not only omission of all direct and indirect inculpations of codefen-dants but also of any statement that could be employed against those defendants once their identity is otherwise established.” 391 U. S., at 143 (emphasis added) (White, J., dissenting).
The author of that opinion today adheres to that interpretation of Bruton. See Cruz v. New York, ante, at 195-196 (White, J., dissenting) (“[A] codefendant’s out-of-court statements implicating the defendant are not only hearsay but also have traditionally been viewed with special suspicion. . . . Bruton held that where the defendant has not himself confessed, there is too great a chance that the jury would rely on the codefendant’s confession”).
See Memorandum from David L. Cook, Administrative Office of the United States Courts, to Supreme Court Library (Feb. 20,1987) (available in Clerk of Court’s case file).
See Memorandum from David L. Cook, Administrative Office of the United States Courts, to Supreme Court Library (Mar. 25,1987) (available in Clerk of Court’s case file) (establishing that figures cited in Memoran*219dum of February 20, 1987, cited ante, at 209, carry no information whatever about the number of multiple-defendant trials in which a codefen-dant’s confession was offered or admitted).
The Court expresses an apparently deep-seated fear that an evenhanded application of Bruton would jeopardize the use of joint trials. This proposition rests on the unsupported assumption that the number of powerfully incriminating confessions that do not name the defendant is too large to be evaluated on a case-by-case basis. The Court then proceeds to the ostensible administrative outrages of the separate trials that would be necessary, contending that it would be unwise to compel prosecutors to “bring separate proceedings, presenting the same evidence again and again, requiring victims and witnesses to repeat the inconvenience (and sometimes trauma) of testifying, and randomly favoring the last-tried defendants who have the advantage of knowing the prosecution’s case beforehand.” Ante, at 210. This speculation also floats unattached to any anchor of reality. Since the likelihood that more than one of the defendants in a joint trial will have confessed is fairly remote, the prospect of “presenting the same evidence again and again” is nothing but a rhetorical flourish. At worst, in the typical case, two trials may be required, one for the confessing defendant and another for the nonconfessing defendant or defendants. And even in that category, presumably most confessing defendants are likely candidates for plea bargaining.
Except for Williams’ confession, and the prosecutor’s closing argument that will be separately considered on remand, there was a paucity of other evidence connecting respondent with the plan discussed in the car on the way to the victims’ home. The Court of Appeals was thus unquestionably correct in concluding that the violation of the Confrontation Clause in this case was not harmless error.