concurring in part and dissenting in part.
As I understand the Court’s opinion, it distinguishes among three categories of evidence related to insanity: (1) fact-related evidence as to the defendant’s specific state of mind at the time of the crime, e. g., evidence that shows he *780thought the policeman was not a human being; (2) expert opinion evidence that the defendant suffered from a mental disease that would have affected his capacity to form an intent to kill a policeman, e. g., that he suffers from a disease of a kind where powerful voices command the sufferer to kill; and (3) expert opinion evidence that the defendant was legally insane, e. g., evidence that he did not know right from wrong. Ante, at 757-759.
I agree with the Court’s basic categorization. I also agree that the Constitution permits a State to provide for consideration of the second and third types of evidence solely in conjunction with the insanity defense. A State might reasonably fear that, without such a rule, the types of evidence as to intent would become confused in the jury’s mind, indeed that in some cases the insanity question would displace the intent question as the parties litigate both simultaneously.
Nonetheless, I believe the distinction among these kinds of evidence will be unclear in some cases. And though I accept the majority’s reading of the record, I remain concerned as to whether the lower courts, in setting forth and applying State v. Mott, 187 Ariz. 536, 931 P. 2d 1046, cert. denied, 520 U. S. 1234 (1997), focused with sufficient directness and precision upon the distinction.
Consequently, I would remand this case so that Arizona’s courts can determine whether Arizona law, as set forth in Mott and other cases, is consistent with the distinction the Court draws and whether the trial court so applied Arizona law here. I would also reserve the question (as I believe the Court has done) as to the burden of persuasion in a case where the defendant produces sufficient evidence of the second kind as to raise a reasonable doubt suggesting that he suffered from a mental illness so severe as to prevent him from forming any relevant intent at all.
For this reason, I dissent only from Parts III-B and III-C of the Court’s opinion and the ultimate disposition of this case, and I join the remainder.
*781Justice Kennedy, with whom Justice Stevens and Justice Ginsburg join, dissenting.
In my submission the Court is incorrect in holding that Arizona may convict petitioner Eric Clark of first-degree murder for the intentional or knowing killing of a police officer when Clark was not permitted to introduce critical and reliable evidence showing he did not have that intent or knowledge. The Court is wrong, too, when it concludes the issue cannot be reached because of an error by Clark’s counsel. Its reasons and conclusions lead me to file this respectful dissent.
Since I would reverse the judgment of the Arizona Court of Appeals on this ground, and the Arizona courts might well alter their interpretation of the State’s criminal responsibility statute were my rationale to prevail, it is unnecessary for me to address the argument that Arizona’s definition of insanity violates due process.
I
Clark claims that the trial court erred in refusing to consider evidence of his chronic paranoid schizophrenia in deciding whether he possessed the knowledge or intent required for first-degree murder. Seizing upon a theory invented here by the Court itself, the Court narrows Clark’s claim so he cannot raise the point everyone else thought was involved in the case. The Court says the only issue before us is whether there is a right to introduce mental-disease evidence or capacity evidence, not a right to introduce observation evidence. See ante, at 756-765. This restructured evidentiary universe, with no convincing authority to support it, is unworkable on its own terms. Even were that not so, however, the Court’s tripartite structure is something not addressed by the state trial court, the state appellate court, counsel on either side in those proceedings, or the briefs the parties filed with us. The Court refuses to consider the key part of Clark’s claim because his counsel did *782not predict the Court’s own invention. It is unrealistic, and most unfair, to hold that Clark’s counsel erred in failing to anticipate so novel an approach. If the Court is to insist on its approach, at a minimum the case should be remanded to determine whether Clark is bound by his counsel’s purported waiver.
The Court’s error, of course, has significance beyond this case. It adopts an evidentiary framework that, in my view, will be unworkable in many cases. The Court classifies Clark’s behavior and expressed beliefs as observation evidence but insists that its description by experts must be mental-disease evidence or capacity evidence. See ante, at 757-759. These categories break down quickly when it is understood how the testimony would apply to the question of intent and knowledge at issue here. The most common type of schizophrenia, and the one Clark suffered from, is paranoid schizophrenia. See P. Berner et al., Diagnostic Criteria for Functional Psychoses 37 (2d ed. 1992). The existence of this functional psychosis is beyond dispute, but that does not mean the lay witness understands it or that a disputed issue of fact concerning its effect in a particular instance is not something for the expert to address. Common symptoms of the condition are delusions accompanied by hallucinations, often of the auditory type, which can cause disturbances of perception. Ibid. Clark’s expert testified that people with schizophrenia often play radios loudly to drown out the voices in their heads. See App. 32. Clark’s attorney argued to the trial court that this, rather than a desire to lure a policeman to the scene, explained Clark’s behavior just before the killing. Id., at 294-295. The observation that schizophrenics play radios loudly is a fact regarding behavior, but it is only a relevant fact if Clark has schizophrenia.
Even if this evidence were, to use the Court’s term, mental-disease evidence, because it relies on an expert opinion, what would happen if the expert simply were to testify, without mentioning schizophrenia, that people with Clark’s *783symptoms often play the radio loudly? This seems to be factual evidence, as the term is defined by the Court, yet it differs from mental-disease evidence only in forcing the witness to pretend that no one has yet come up with a way to classify the set of symptoms being described. More generally, the opinion that Clark had paranoid schizophrenia—an opinion shared by experts for both the prosecution and defense—bears on efforts to determine, as a factual matter, whether he knew he was killing a police officer. The psychiatrist’s explanation of Clark’s condition was essential to understanding how he processes sensory data and therefore to deciding what information was in his mind at the time of the shooting. Simply put, knowledge relies on cognition, and cognition can be affected by schizophrenia. See American Psychiatric Association, Diagnostic and Statistical Manual of Mental Disorders 299 (4th ed. text rev. 2000) (“The characteristic symptoms of Schizophrenia involve a range of cognitive and emotional dysfunctions that include perception”); ibid. (Symptoms include delusions, which are “erroneous beliefs that usually involve a misinterpretation of perceptions or experiences”). The mental-disease evidence at trial was also intertwined with the observation evidence because it lent needed credibility. Clark’s parents and friends testified Clark thought the people in his town were aliens trying to kill him. These claims might not be believable without a psychiatrist confirming the story based on his experience with people who have exhibited similar behaviors. It makes little sense to divorce the observation evidence from the explanation that makes it comprehensible.
Assuming the Court’s tripartite structure were feasible, the Court is incorrect when it narrows Clark’s claim to exclude any concern about observation evidence. In deciding Clark’s counsel failed to raise this issue, the Court relies on a series of perceived ambiguities regarding how the claim fits within the Court’s own categories. See ante, at 761-765. The Court cites no precedent for construing these ambiguities against the claimant and no prudential reason for ignor*784ing the breadth of Clark’s claim. It is particularly surprising that the Court does so to the detriment of a criminal defendant asserting the fundamental challenge that the trier of fact refused to consider critical evidence showing he is innocent of the crime charged.
The alleged ambiguities are, in any event, illusory. The evidence at trial addressed more than the question of general incapacity or opinions regarding mental illness; it went further, as it included so-called observation evidence relevant to Clark’s mental state at the moment he shot the officer. There was testimony, for example, that Clark thought the people in his town, particularly government officials, were not human beings but aliens who were trying to kill him. See App. 119-121, 131-132, 192-197, 249-256; Tr. of Bench Trial in No. CR 2000-538, pp. 110-112,131-132,136, 226-228 (Aug. 20, 2003); id., at 24-25, 59-60 (Aug. 21, 2003). The Court recognizes the existence of this essential observation evidence. See ante, at 757-759.
The Court holds, nonetheless, that “we cannot be sure” whether the trial court failed to consider this evidence. Ante, at 764-765. It is true the trial court ruling was not perfectly clear. Its language does strongly suggest, though, that it did not consider any of this testimony in deciding whether Clark had the knowledge or intent required for first-degree murder. After recognizing that “much of the evidence that [the defense is] going to be submitting, in fact all of it, as far as I know ... that has to do with the insanity could also arguably be made ... as to form and intent and his capacity for the intent,” the court concluded “we will be focusing, as far as I’m concerned, strictly on the insanity defense.” App. 9. In announcing its verdict, the trial court did not mention any of the mental-illness evidence, observation or otherwise, in deciding Clark’s guilt. Id., at 331-335. The most reasonable assumption, then, would seem to be that the trial court did not consider it, and the Court does not hold otherwise. See ante, at 760-761.
*785Clark’s objection to this refusal by the trier of fact to consider the evidence as it bore on his key defense was made at all stages of the proceeding. In his post-trial motion to vacate the judgment, Clark argued that “prohibiting consideration of any evidence reflecting upon a mentally ill criminal defendant’s ability to form the necessary mens rea violates due process.” Record, Doc. 406, p. 8. Clark pressed the same argument in the Arizona Court of Appeals. See Appellant’s Opening Brief in No. 1CA-CR-03-0851 etc., pp. 46-52 (hereinafter Appellant’s Opening Brief). He also noted that the trial judge had erred in refusing to consider nonexpert testimony—presumably what the Court would call observation evidence—on Clark’s mental illness. Id,., at 47-48 (“The trial court therefore violated [Clark’s] right to present a defense because [the] court refused to consider any evidence, including the multiple testimonials of lay witnesses ... in deciding whether he could form the requisite mens rea”). The appeals court decided the issue on the merits, holding that the trial court was correct not to consider the evidence of mental illness in determining whether Clark had the mens rea for first-degree murder. See App. 351-353. It offered no distinction at all between observation or mental-disease evidence.
Notwithstanding the appeals court’s decision, the Court states that the issue was not clearly presented to the state courts. See ante, at 762-765. According to the Court, Clark only raised an objection based on State v. Mott, 187 Ariz. 536, 931 P. 2d 1046 (1997), cert. denied, 520 U. S. 1234 (1997), see ante, at 762-765, and Mott’s holding was limited to the exclusion of mental-disease and capacity evidence, see ante, at 760. The Court is incorrect, and on both counts.
First, Clark’s claim goes well beyond an objection to Mott. In fact, he specifically attempted to distinguish Mott by noting that the trial court in this case refused to consider all evidence of mental illness. See Record, Doc. 406, at 8; see *786also Appellant’s Opening Brief 48. The Court notices these arguments but criticizes Clark’s counsel for not being specific about the observation evidence he wanted the trial court to consider. See ante, at 763. There was no reason, though, for Clark’s counsel to believe additional specificity was required, since there was no evident distinction in Arizona law between observation evidence and mental-disease testimony.
Second, Mott’s holding was not restricted to mental-disease evidence. The Arizona Supreme Court did not refer to any distinction between observation and mental-disease evidence, or lay and expert testimony. Its holding was stated in broad terms: “Arizona does not allow evidence of a defendant’s mental disorder short of insanity either as an affirmative defense or to negate the mens rea element of a crime.” 187 Ariz., at 541, 931 P. 2d, at 1051; see id., at 540, 931 P. 2d, at 1050 (“The legislature’s decision . . . evidences its rejection of the use of psychological testimony to challenge the mens rea element of a crime”). The Court attempts to divine a fact/opinion distinction in Mott based on Mott’s distinguishing a case, State v. Christensen, 129 Ariz. 32, 628 P. 2d 580 (1981), where evidence about behavioral tendencies was deemed admissible. See ante, at 760. Christensen, though, also addressed an expert opinion; the difference was that the evidence there concerned a “character trait of acting reflexively in response to stress,” not a mental illness. Mott, supra, at 544, 931 P. 2d, at 1054. Since the Court recognizes the Arizona Court of Appeals relied on Mott, the expansive rule of exclusion in Mott— without any suggestion of a limitation depending on the kind of evidence—should suffice for us to reach the so-called observation-evidence issue. Even if, as the Court contends, see ante, at 760, Mott is limited to expert testimony, the Court’s categories still do not properly interpret Mott, because the Court’s own definition of observation evidence includes some expert testimony, see ante, at 757-758.
It makes no difference that in the appeals court Clark referred to the issue as inability to form knowledge or intent. *787See Appellant’s Opening Brief 46-52. He did not insist on some vague, general incapacity. He stated, instead, that he “suffered from a major mental illness and was psychotic at the time of the offense.” Id., at 48. Even if Clark’s arguments were insufficient to apprise the state court of the argument, “[o]ur traditional rule is that ‘[ojnce a federal claim is properly presented, a party can make any argument in support of that claim; parties are not limited to the precise arguments they made below.’” Lebron v. National Railroad Passenger Corporation, 513 U. S. 374, 379 (1995) (quoting Yee v. Escondido, 503 U. S. 519, 534 (1992)). The claim is clear. Though it seems to be obscure to this Court, it was understood by the Arizona Court of Appeals, which stated: “Clark argues that the trial court erred in refusing to consider evidence of his mental disease or defect in determining whether he had the requisite mens rea to commit first-degree murder.” App. 351. When the question is what the state court held, it is not instructive for this Court to recast the words the state court used.
The razor-thin distinction the Court draws between evidence being used to show incapacity and evidence being used to show lack of mens rea directly does not identify two different claims. Clark’s single claim, however characterized, involves the use of the same mental-illness evidence to decide whether he had the requisite knowledge or intent. The various ways in which the evidence is relevant in disproving mens rea hardly qualify as separate claims. The new arguments allowed in Lebrón and Yee, by comparison, were far more disconnected from the initial bases for the alleged violations. See Lebron, supra, at 378, 379 (for purposes of showing state action, petitioner could argue that Amtrak was a Government entity even though he argued below only that it was a private entity with close connections to Government entities, because the claim was simply “that Amtrak did not accord him the rights it was obliged to provide by the First Amendment”); Yee, supra, at 534, 535 (petitioners *788could argue that an ordinance constituted a regulatory taking, even though they arguably asserted in the Court of Appeals only a physical taking, because the claim was simply “that the ordinance effects an unconstitutional taking”). If we give this latitude to litigants in civil cases, surely we must do so here. Furthermore, to the extent any ambiguity remains on whether the claim was raised, the proper course is to remand. See Bradshaw v. Richey, 546 U. S. 74, 80 (2005) (per curiam). Unless the state court clearly decides an issue on state-law grounds, which the Court does not contend occurred here, there is no bar to our review of the federal question. See Harris v. Reed, 489 U. S. 255, 261-262 (1989).
Before this Court Clark framed the issue in broad terms that encompass the question whether the evidence of his mental illness should have been considered to show he did not at the time of the offense have the knowledge or intent to shoot a police officer. See Brief for Petitioner i (“Questions Presented for Review (1) Whether Arizona’s blanket exclusion of evidence and refusal to consider mental disease or defect to rebut the state’s evidence on the element of mens rea violated Petitioner’s right to due process under the United States Constitution, Fourteenth Amendment?”), 22 (“Here, the trial court held that under the Mott rule it was obliged to find as a fact that [Clark] knew he was shooting a police officer to death—a necessary factual element of the only form of first degree murder charged against [Clark]— while simultaneously refusing to consider [Clark’s] evidence that an acute episode of his chronic paranoid schizophrenic illness prevented him from actually having that knowledge” (emphasis deleted)), 31-32 (the Arizona courts erred in holding Clark “could be punished as though he had this knowledge and' intent although he may not in fact have had either”); Reply Brief for Petitioner 3 (challenging the trial judge’s refusal “to give any consideration to the mental-illness evidence in making his factual findings as to whether *789[Clark] did or did not act with the state of mind required for a first-degree murder conviction”). An entire section of Clark’s opening brief argues that the evidence of mental illness should have been considered to rebut the prosecution’s inference of knowledge or intent from the factual circumstances of the crime. See Brief for Petitioner 13-21. This line of argument concerns facts of behavior and amounts to more than a claim of general incapacity.
Clark seeks resolution of issues that can be complex and somewhat overlapping. In the end, however, we must decide whether he had the right to introduce evidence showing he lacked the intent or knowledge the statute itself sets forth in describing a basic element of the crime. Clark has preserved this issue at all stages, including in this Court.
II
Clark was charged with first-degree murder for the shooting of Officer Jeffrey Moritz. “A person commits first-degree murder if,” as relevant here, “[intending or knowing that the person’s conduct will cause death to a law enforcement officer, the person causes the death of a law enforcement officer who is in the line of duty.” Ariz. Rev. Stat. Ann. § 13-1105(A)(3) (West Supp. 2005). Clark challenges the trial court’s refusal to consider any evidence of mental illness, from lay or expert testimony, in determining whether he acted with the knowledge or intent element of the crime. See App. 9; see also Mott, 187 Ariz., at 541, 931 P. 2d, at 1051.
States have substantial latitude under the Constitution to define rules for the exclusion of evidence and to apply those rules to criminal defendants. See United States v. Scheffer, 523 U. S. 303, 308 (1998). This authority, however, has constitutional limits. “Whether rooted directly in the Due Process Clause of the Fourteenth Amendment or in the Compulsory Process or Confrontation Clauses of the Sixth Amendment, the Constitution guarantees criminal defendants “a meaningful opportunity to present a complete de*790fense.”’” Holmes v. South Carolina, 547 U. S. 319, 324 (2006) (quoting Crane v. Kentucky, 476 U. S. 683, 690 (1986), in turn quoting California v. Trombetta, 467 U. S. 479, 485 (1984)). “This right is abridged by evidence rules that ‘infring[e] upon a weighty interest of the accused’ and are ‘“arbitrary” or “disproportionate to the purposes they are designed to serve.” ’ ” Holmes, supra, at 324 (quoting Scheffer, supra, at 308, in turn citing and quoting Rock v. Arkansas, 483 U. S. 44, 58, 56 (1987)).
The central theory of Clark’s defense was that his schizophrenia made him delusional. He lived in a universe where the delusions were so dominant, the theory was, that he had no intent to shoot a police officer or knowledge he was doing so. It is one thing to say he acted with intent or knowledge to pull the trigger. It is quite another to say he pulled the trigger to kill someone he knew to be a human being and a police officer. If the trier of fact were to find Clark’s evidence sufficient to discount the case made by the State, which has the burden to prove knowledge or intent as an element of the offense, Clark would not be guilty of first-degree murder under Arizona law.
The Court attempts to diminish Clark’s interest by treating mental-illness evidence as concerning only “judgment,” rather than fact. Ante, at 776-777. This view appears to derive from the Court’s characterization of Clark’s claim as raising only general incapacity. See ibid. This is wrong for the reasons already discussed. It fails to recognize, moreover, the meaning of the offense element in question here. The mens rea element of intent or knowledge may, at some level, comprise certain moral choices, but it rests in the first instance on a factual determination. That is the fact Clark sought to put in issue. Either Clark knew he was killing a police officer or he did not.
The issue is not, as the Court insists, whether Clark’s mental illness acts as an “excuse from customary criminal responsibility,” ante, at 771, but whether his mental illness, as *791a factual matter, made him unaware that he was shooting a police officer. If it did, Clark needs no excuse, as then he did not commit the crime as Arizona defines it. For the elements of first-degree murder, where the question is knowledge of particular facts—that one is killing a police officer— the determination depends not on moral responsibility but on empirical fact. Clark’s evidence of mental illness had a direct and substantial bearing upon what he knew, or thought he knew, to be the facts when he pulled the trigger; this lay at the heart of the matter.
The trial court’s exclusion was all the more severe because it barred from consideration on the issue of mens rea all this evidence, from any source, thus preventing Clark from showing he did not commit the crime as defined by Arizona law. Quite apart from due process principles, we have held that a bar of this sort can be inconsistent with the Confrontation Clause. See Delaware v. Van Arsdall, 475 U. S. 673 (1986). In Van Arsdall the Court held a state court erred in making a ruling that “prohibited all inquiry into” an event. Id., at 679. At issue was a line of defense questioning designed to show the bias of a prosecution witness. In the instant case the ruling in question bars from consideration all testimony from all witnesses necessary to present the argument that was central to the whole case for the defense: a challenge to the State’s own proof on an element of the crime. The Due Process and Compulsory Process Clauses, and not the Confrontation Clause, may be the controlling standard; but the disability imposed on the accused is every bit as substantial and pervasive here as it was in Van Arsdall.
Arizona’s rule is problematic because it excludes evidence no matter how credible and material it may be in disproving an element of the offense. The Court’s cases have noted the potential arbitrariness of per se exclusions and, on this rationale, have invalidated various state prohibitions. See Holmes, supra, at 329 (rule excluding, in certain cases, evidence that a third party may have committed the crime *792“even if that evidence, if viewed independently, would have great probative value and even if it would not pose an undue risk of harassment, prejudice, or confusion of the issues”); Rock, supra, at 56 (rule excluding all hypnotically refreshed testimony “operates to the detriment of any defendant who undergoes,hypnosis, without regard to the reasons for it, the circumstances under which it took place, or any independent verification of the information it produced”); Washington v. Texas, 388 U. S. 14, 22 (1967) (rule excluding accomplice testimony “prevent[s] whole categories of defense witnesses from testifying on the basis of a priori categories that presume them unworthy of belief”).
This is not to suggest all general rules on the exclusion of certain types of evidence are invalid. If the rule does not substantially burden the defense, then it is likely permissible. See Scheffer, 523 U. S., at 316-317 (upholding exclusion of polygraph evidence in part because this rule “does not implicate any significant interest of the accused”); id., at 318 (Kennedy, J., concurring in part and concurring in judgment) (“[S]ome later case might present a more compelling case for introduction of the testimony than this one does”). Where, however, the burden is substantial, the State must present a valid reason for its per se evidentiary rule.
In the instant case Arizona’s proposed reasons are insufficient to support its categorical exclusion. While the State contends that testimony regarding mental illness may be too incredible or speculative for the jury to consider, this does not explain why the exclusion applies in all cases to all evidence of mental illness. “A State’s legitimate interest in barring unreliable evidence does not extend to per se exclusions that may be reliable in an individual case.” Rock, supra, at 61. States have certain discretion to bar unreliable or speculative testimony and to adopt rules to ensure the reliability of expert testimony. Arizona has done so, and there is no reason to believe its rules are insufficient to avoid *793speculative evidence of mental illness. See Ariz. Rules Evid. 403, 702 (2006). This is particularly true because Arizona applies its usual case-by-case approach to permit admission of evidence of mental illness for a variety of other purposes. See, e. g., State v. Lindsey, 149 Ariz. 472, 474-475, 720 P. 2d 73, 75-76 (1986) (en banc) (psychological characteristics of molestation victims); State v. Hamilton, 177 Ariz. 403, 408-410, 868 P. 2d 986, 991-993 (App. 1993) (psychological evidence of child abuse accommodation syndrome); Horan v. Industrial Common of Ariz., 167 Ariz. 322, 325-326, 806 P. 2d 911, 914-915 (App. 1991) (psychiatric testimony regarding neurological deficits).
The risk of jury confusion also fails to justify the rule. The State defends its rule as a means to avoid the complexities of determining how and to what degree a mental illness affects a person’s mental state. The difficulty of resolving a factual issue, though, does not present a sufficient reason to take evidence away from the jury even when it is crucial for the defense. “We have always trusted juries to sort through complex facts in various areas of law.” United States v. Booker, 543 U. S. 220, 289 (2005) (Stevens, J., dissenting in part). Even were the risk of jury confusion real enough to justify excluding evidence in most cases, this would provide little basis for prohibiting all evidence of mental illness without any inquiry into its likely effect on the jury or its role in deciding the linchpin issue of knowledge and intent. Indeed, Arizona has a rule in place to serve this very purpose. See Rule 403.
Even assuming the reliability and jury-confusion justifications were persuasive in some cases, they would not suffice here. It does not overcome the constitutional objection to say that an evidentiary rule that is reasonable on its face can be applied as well to bar significant defense evidence without any rational basis for doing so. In Van Arsdall, for example, the Court rejected the application of Delaware Rule of Evidence 403, which allows relevant evidence to be excluded *794where its probative value is substantially outweighed by the risk of unfair prejudice or other harms to the trial process. 475 U. S., at 676, and n. 2. While the Rule is well established and designed for a legitimate function, the Constitution prevented an application that deprived the defendant of all inquiry into an important issue. Id., at 679. Other cases have applied this same ease-specific analysis in deciding the legitimacy of an exclusion. See, e. g., Rock, 488 U. S., at 62 (the “circumstances present an argument for admissibility of petitioner’s testimony in this particular case, an argument that must be considered by the trial court”); Chambers v. Mississippi, 410 U. S. 284, 302 (1973) (“In these circumstances, where constitutional rights directly affecting the ascertainment of guilt are implicated, the hearsay rule may not be applied mechanistically to defeat the ends of justice”); cf. Scheffer, supra, at 318 (Kennedy, J., concurring in part and concurring in judgment).
The Court undertakes little analysis of the interests particular to this case. By proceeding in this way it devalues Clark’s constitutional rights. The reliability rationale has minimal applicability here. The Court is correct that many mental diseases are difficult to define and the subject of great debate. See ante, at 774-775. Schizophrenia, however, is a well-documented mental illness, and no one seriously disputes either its definition or its most prominent clinical manifestations. The State’s own expert conceded that Clark had paranoid schizophrenia and was actively psychotic at the time of the killing. See App. 254-257. The jury-confusion rationale, if it is at all applicable here, is the result of the Court’s own insistence on conflating the insanity defense and the question of intent. Considered on its own terms, the issue of intent and knowledge is a straightforward factual question. A trier of fact is quite capable of weighing defense testimony and then determining whether the accused did or did not intend to kill or knowingly kill a human being who was a police officer. True, the issue can be diffi*795cult to decide in particular instances, but no more so than many matters juries must confront.
The Court says mental-illness evidence “can easily mislead,” ante, at 776, and may “tel[l] us little or nothing about the ability of the defendant to form mens rea,” ante, at 775. These generalities do not, however, show how relevant or misleading the evidence in this case would be (or explain why Arizona Rule of Evidence 403 is insufficient for weighing these factors). As explained above, the evidence of Clark’s mental illness bears directly on mens rea, for it suggests Clark may not have known he was killing a human being. It is striking that while the Court discusses at length the likelihood of misjudgment from placing too much emphasis on evidence of mental illness, see ante, at 773-778, it ignores the risk of misjudging an innocent man guilty from refusing to consider this highly relevant evidence at all. Clark’s expert, it is true, said no one could know exactly what was on Clark’s mind at the time of the shooting. See ante, at 777. The expert testified extensively, however, about the effect of Clark’s delusions on his perceptions of the world around him, and about whether Clark’s behavior around the time of the shooting was consistent with delusional thinking. This testimony was relevant to determining whether Clark knew he was killing a human being. It also bolstered the testimony of lay witnesses, none of which was deemed unreliable or misleading by the state courts.
For the same reasons, the Court errs in seeking support from the American Psychiatric Association’s statement that a psychiatrist may be justifiably reluctant to reach legal conclusions regarding the defendant’s mental state. See ante, at 777-778. In this very ease, the American Psychiatric Association made clear that psychiatric evidence plays a crucial role regardless of whether the psychiatrist testifies on the ultimate issue: “Expert evidence of mental disorders, presented by qualified professionals and subject to adversarial testing, is both relevant to the mental-state issues raised by *796mens rea requirements and reliable.... Such evidence could not be condemned wholesale without unsettling the legal system’s central reliance on such evidence.” Brief for American Psychiatric Association et al. as Amici Curiae 15.
Contrary to the Court’s suggestion, see ante, at 776, the fact that the state and defense experts drew different conclusions about the effect of Clark’s mental illness on his mental state only made Clark’s evidence contested; it did not make the evidence irrelevant or misleading. The trial court was capable of evaluating the competing conclusions, as factfinders do in countless cases where there is a dispute among witnesses. In fact, the potential to mislead will be far greater under the Court’s new evidentiary system, where jurors will receive observation evidence without the necessary explanation from experts.
The fact that mental-illness evidence may be considered in deciding criminal responsibility does not compensate for its exclusion from consideration on the mens rea elements of the crime. Cf. ante, at 773-774. The evidence addresses different issues in the two instances. Criminal responsibility involves an inquiry into whether the defendant knew right from wrong, not whether he had the mens rea elements of the offense. While there may be overlap between the two issues, “the existence or nonexistence of legal insanity bears no necessary relationship to the existence or nonexistence of the required mental elements of the crime.” Mullaney v. Wilbur, 421 U. S. 684, 706 (1975) (Rehnquist, J., concurring).
Even if the analyses were equivalent, there is a different burden of proof for insanity than there is for mens rea. Arizona requires the defendant to prove his insanity by clear and convincing evidence. See Ariz. Rev. Stat. Ann. § 13— 502(C) (West 2001). The prosecution, however, must prove all elements of the offense beyond a reasonable doubt. See Mullaney, supra, at 703-704; In re Winship, 397 U. S. 358, 364 (1970). The shift in the burden on the criminal responsibility issue, while permissible under our precedent, see Le*797land v. Oregon, 343 U. S. 790 (1952), cannot be applied to the question of intent or knowledge without relieving the State of its responsibility to establish this element of the offense. See Sandstrom v. Montana, 442 U. S. 510, 524 (1979) (jury instruction that had the effect of placing the burden on the defendant to disprove that he had the requisite mental state violates due process). While evidentiary rules do not generally shift the burden impermissibly, where there is a right to have evidence considered on an element of the offense, the right is not respected by allowing the evidence to come in only on an issue for which the defendant bears the burden of proof. See Cool v. United States, 409 U. S. 100, 103 (1972) (per curiam) (jury instruction that allowed jury to consider accomplice’s testimony only if it was true beyond a reasonable doubt “places an improper burden on the defense and allows the jury to convict despite its failure to find guilt beyond a reasonable doubt”); Martin v. Ohio, 480 U. S. 228, 233-234 (1987) (State can shift the burden on a claim of self-defense, but if the jury were disallowed from considering self-defense evidence for purposes of deciding the elements of the offense, it “would relieve the State of its burden and plainly run afoul of Winship’s mandate”). By viewing the Arizona rule as creating merely a “presumption of sanity (or capacity or responsibility),” ante, at 771, rather than a presumption that the mens rea elements were not affected by mental illness, the Court fails to appreciate the implications for Winship.
The State attempts to sidestep the evidentiary issue entirely by claiming that its mental-illness exclusion simply alters one element of the crime. The evidentiary rule at issue here, however, cannot be considered a valid redefinition of the offense. Under the State’s logic, a person would be guilty of first-degree murder if he knowingly or intentionally killed a police officer or committed the killing under circumstances that would show knowledge or intent but for the defendant’s mental illness. To begin with, Arizona law does *798not say this. And if it did, it would be impermissible. States have substantial discretion in defining criminal offenses. In some instances they may provide that the accused has the burden of persuasion with respect to affirmative defenses. See Patterson v. New York, 432 U. S. 197, 210 (1977). “But there are obviously constitutional limits beyond which the States may not go in this regard.” Ibid. If it were otherwise, States could label all evidentiary exclusions as redefinitions and so evade constitutional requirements. There is no rational basis, furthermore, for criminally punishing a person who commits a killing without knowledge or intent only if that person has a mental illness. Cf. Robinson v. California, 370 U. S. 660, 666 (1962). The State attempts to bring the instant case within the ambit of Montana v. Egelhoff, 518 U. S. 37 (1996); but in Egelhoff the excluded evidence concerned voluntary intoxication, for which a person can be held responsible. Viewed either as an evidentiary rule or a redefinition of the offense, it was upheld because it “comports with and implements society’s moral perception that one who has voluntarily impaired his own faculties should be responsible for the consequences.” Id., at 50 (plurality opinion). An involuntary mental illness does not implicate this justification.
Future dangerousness is not, as the Court appears to conclude, see ante, at 778-779, n. 45, a rational basis for convicting mentally ill individuals of crimes they did not commit. Civil commitment proceedings can ensure that individuals who present a danger to themselves or others receive proper treatment without unfairly treating them as criminals. The State presents no evidence to the contrary, and the Court ought not to imply otherwise.
The State gains little support from Fisher v. United States, 328 U. S. 463 (1946). There the defendant requested an instruction from the trial court that the jury consider his mental deficiencies in determining his capacity for premeditation and deliberation. Id., at 470. The Court noted that *799“[i]n view of the status of the defense of partial responsibility in the District and the nation no contention is or could be made of the denial of due process.” Id., at 466. This dictum may be attributable to the fact that the cases recognizing a defendant’s evidentiary rights and the prosecution’s duty to prove all elements beyond a reasonable doubt were still decades away. It may also reflect the fact that the jury instructions as given did seem to allow the jury to consider evidence of mental deficiency if it disproved the elements of the offense. See id., at 467, n. 3 (The jury instructions stated, “Tt is further contended that even if sane and responsible, there was no deliberate intent to kill, nor in fact any actual intent to kill. Therefore if not guilty by reason of insanity, the defendant at most is guilty only of second degree murder or manslaughter’ ”). Even further ambiguity comes from the fact that the defense in Fisher concerned a claim that the petitioner was “mentally somewhat below the average” with a “psychopathic personality” of aggression. Id., at 467. This general claim of mental deficiencies was relevant to the “theory of partial responsibility,” id., at 470, he wanted the jury to consider. Unlike the mental illness here, though, which concerns inadequacy of perception and information processing, the petitioner’s claim may not have been relevant to mens rea unless mens rea were redefined to include an element of responsibility. Fisher’s language, then, does not control this case.
While Arizona’s rule is not unique, either historically or in contemporary practice, this fact does not dispose of Clark’s constitutional argument. To the extent Fisher may have suggested the contrary, subsequent cases make clear that while the existence of the rule in some jurisdictions is a significant factor to consider, see Egelhoff, supra, at 43 (plurality opinion), it is not dispositive for evaluation of a claim that the accused was foreclosed from introducing evidence crucial to the defense. The evidentiary exclusion of accomplice testimony the Court invalidated in Washington was, in fact, *800well established. See 388 U. S., at 21-22. The exclusion of hypnotically refreshed testimony likewise had some support when the Court held it unconstitutional as applied to a defendant’s own testimony. Rock, 483 U. S., at 57. While 13 States still impose significant restrictions on the use of mental-illness evidence to negate mens rea, a substantial majority of the States currently allow it. Brief for United States as Amicus Curiae 22-23, and n. 13. The fact that a reasonable number of States restrict this evidence weighs into the analysis, but applying the rule as a per se bar, as Arizona does, is so plainly unreasonable that it cannot be sustained.
Putting aside the lack of any legitimate state interest for application of the rule in this case, its irrationality is apparent when considering the evidence that is allowed. See Washington, supra, at 22 (“The absurdity of the rule is amply demonstrated by the exceptions that have been made to it”). Arizona permits the defendant to introduce, for example, evidence of “behavioral tendencies” to show he did not have the required mental state. See Mott, 187 Ariz., at 544, 931 P. 2d, at 1054; Christensen, 129 Ariz., at 35-36, 628 P. 2d, at 583-584. While defining mental illness is a difficult matter, the State seems to exclude the evidence one would think most reliable by allowing unexplained and uncategorized tendencies to be introduced while excluding relatively well-understood psychiatric testimony regarding well-documented mental illnesses. It is unclear, moreover, what would have happened in this case had the defendant wanted to testify that he thought Officer Moritz was an alien. If disallowed, it would be tantamount to barring Clark from testifying on his behalf to explain his own actions. If allowed, then Arizona’s rule would simply prohibit the corroboration necessary to make sense of Clark’s explanation. In sum, the rule forces the jury to decide guilt in a fictional world with undefined and unexplained behaviors but without mental illness. This rule has no rational justification and *801imposes a significant burden upon a straightforward defense: He did not commit the crime with which he was charged.
These are the reasons for my respectful dissent.