with whom Justice Stevens joins, concurring in part and dissenting in part.
I agree with the Court that the evidence of battered child syndrome was relevant. The State had to prove that Mark *76McGuire intended to kill his daughter, and the evidence that Tori was a battered child was probative of causation and intent. I therefore join Part I of the Court's opinion.
I do not join Part II of the opinion because I think there is a reasonable likelihood that the jury misapplied the prior acts instruction. The trial court instructed the jury that evidence of Tori’s prior injuries had been admitted to show that McGuire had committed offenses similar to that for which he was on trial, and that, if the jury found a “clear connection” between the prior offenses and the charged offense, “it may be logically concluded that if the Defendant committed other offenses, he also committed the crime charged in this case.” App. 41. In my view, the instruction encouraged the jury to assume that McGuire had inflicted the prior injuries and then directed the jury to conclude that the prior abuser was the murderer. Because the instruction may have relieved the State of its burden of proving the identity of Tori’s murderer beyond a reasonable doubt, I would hold that the instruction was error and remand to the Court of Appeals for a determination of whether that error was harmless.
The fact that a 6-month-old child was repeatedly beaten in the course of her short life is so horrifying that a trial court should take special care to inform the jury as to the significance of that evidence. As the Court notes, the demonstration of battered child syndrome is relevant because it “ ‘indicates that a child found with [serious, repeated injuries] has not suffered those injuries by accidental means,’ ” ante, at 68 (quoting People v. Jackson, 18 Cal. App. 3d 504, 507; 95 Cal. Rptr. 919, 921 (1971)). L therefore agree that proof of Tori’s battered child status, although “not linked by any direct evidence to McGuire,” was properly admitted because “the evidence demonstrated that Tori’s death was the result of an intentional act by someone, and not an accident.” Ante, at 69. Precisely because the relevance of battered child syndrome is not tied to the identity of the abuser, however, I *77believe that a jury instruction clarifying the limited probative value of that evidence was required.
Instead of an instruction limiting the use of evidence of Tori’s prior injuries, the trial judge gave an instruction limiting the use of evidence of McGuire’s prior bad acts. In so doing, the trial judge himself appears to have assumed that the prior injuries could be attributed to McGuire. The judge told the jury that “[ejvidence has been introduced for the purpose of showing that the Defendant committed acts similar to those constituting a crime other than that for which he is on trial.” App. 40. The State concedes that this category of evidence encompasses both the acts for which McGuire was positively identified as the actor (carrying the child by one arm and roughly pinching her cheeks) and the far more brutal acts for which no actor was identified (the fractured ribs and the rectal tearing). The grouping of these two distinct classes of evidence created a reasonable likelihood that the jury would believe that McGuire had been identified — at least in the eyes of the trial judge — as the prior abuser.
The trial court’s error in implying that McGuire had been identified as the prior abuser was compounded by its further instruction that, if the jury found a “clear connection” between the prior offénses and the charged offense, “it may be logically concluded that if the Defendant committed other offenses, he also committed the crime charged in this case.” The Court finds it “likely” that the jury understood the instruction to mean that “if it found a ‘clear connection’ between the prior injuries and the instant injuries, and if it found that McGuire had committed the prior injuries, then it could use that fact in determining that McGuire committed the crime charged.” Ante, at 75. In my view, there is a reasonable likelihood that the jury did not understand this single sentence to establish a two-step process.
The jury was instructed to “consider” the evidence that McGuire had “committed acts similar” to the crime charged *78and to “determin[e]” whether there was a “clear connection” between these prior acts and the ones that resulted in Tori’s death. App. 40, 41. The trial court did not instruct the jury that it must first “determine” whether McGuire had in fact inflicted the prior injuries. The part of the instruction relied upon by the Court — “it may be logically concluded that if the defendant committed other offenses, he also committed the crime charged in this case” — does not make clear that it is the jury’s role to ascertain whether McGuire was the perpetrator of the prior abuse. Rather, coming as it does in the middle of what appears to be a conclusion of law, it is reasonably likely that the jury understood that such a determination had already been made and that its role was merely to determine if there was a “clear connection” between Tori’s prior injuries and the injuries that killed her.
Although we “have defined the category of infractions that violate ‘fundamental fairness’ very narrowly,” Dowling v. United States, 493 U. S. 342, 352 (1990), it is well established that the fundamental fairness guarantee of the Due Process Clause requires the prosecution to prove beyond a reasonable doubt every element of the offense. In re Winship, 397 U. S. 358, 364 (1970); McMillan v. Pennsylvania, 477 U. S. 79, 85 (1986). This constitutional principle “prohibits the State from using evidentiary presumptions in a jury charge that have the effect of relieving the State of its burden of persuasion beyond a reasonable doubt of every essential element of a crime.” Francis v. Franklin, 471 U. S. 307, 313 (1985); Sandstrom v. Montana, 442 U. S. 510 (1979). Thus, we have held that mandatory presumptions violate the Due Process Clause if they relieve the State of the burden of persuasion on an element of the offense. Patterson v. New York, 432 U. S. 197, 215 (1977); Sandstrom, supra, at 520-524. By contrast, a permissive inference is not a violation *79of due process because the State still has the burden of persuading the jury that the suggested conclusion should be inferred based on the predicate facts proved. Ulster County Court v. Allen, 442 U. S. 140, 157-163 (1979).
In this case, the instruction perhaps was intended to posit a permissive inference that whoever had inflicted Tori’s prior injuries was likely to have inflicted the injuries that caused her death. But the trial court did not make clear that the State first had to prove the predicate facts from which the inference was to be drawn. Furthermore, the wording of the instruction is such that the jury may well have assumed that it had no choice but to “logically conclud[e]” that McGuire was the murderer once it found a “clear connection” between the prior injuries and the fatal ones.* Because I cannot say with any confidence that the instruction allowed a mere permissive inference drawing from proven facts, I think the instruction should be treated as a mandatory presumption that may have relieved the State of its burden of proving the identity of Tori’s killer beyond a reasonable doubt.
Had the instruction been clearly worded, I would agree with the Court that there is sufficient circumstantial evidence in the record to support a finding that McGuire was the perpetrator of the prior injuries. After all, as the Court points out, “[t]he proof of battered child syndrome itself narrowed the group of possible perpetrators to McGuire and his wife, because they were the only two people regularly caring for Tori.” Ante, at 74. In this case, however, it is important to remember that the other person regularly caring for Tori — Daisy McGuire — took the stand and testified, under a grant of immunity, that she was the one who inflicted the fatal injuries on the night of July 7, 1981.
*80McGuire’s jury deliberated for three days before returning a verdict of guilty. Any evaluation of the jury instruction must be conducted against the background of Daisy McGuire’s surprise testimony and the dilemma it so clearly posed for the jury. In my view, the jury instruction on similar acts was so “ambiguous,” ante, at 72, that there was a reasonable likelihood that the jury was encouraged to make assumptions and conclusions about the identity of Tori’s murderer that relieved the State of having to prove that element of the offense beyond a reasonable doubt. In cases where the Court has found that jury instructions included mandatory presumptions inconsistent with the guarantees of the Due Process Clause, the Court has remanded to determine whether the erroneous instruction was harmless, which is the course that should be followed here. See, e. g., Sandstrom, supra, at 526-527; Rose v. Clark, 478 U. S. 570, 584 (1986); Carella v. California, 491 U. S. 263, 266-267 (1989) (per curiam).
Although not dispositive, it is worth noting that California’s model jury instructions on “evidence of other crimes” has since been revised to eliminate the phrase “so that it may be logically concluded.” See 1 California Jury Instructions, Criminal 2.50 (5th ed. 1987).