In January 2008, a Wisconsin appeals court granted a new trial to Audrey Edmunds, a 45-year-old woman who had been sentenced in 1995 to 18 years in prison for murdering Natalie Beard, an infant in her care. The ruling was significant, because medical experts said Beard died as a result of Shaken Baby Syndrome (SBS), a diagnosis that grew increasingly common in the late 1980s and early 1990s. The Wisconsin appellate court was the first in the country to recognize increasing doubts about the reliability of SBS diagnoses.

The phrase shaken baby syndrome entered the pop culture lexicon in 1997, when British au pair Louise Woodward was convicted of involuntary manslaughter in the death of Massachusetts infant Matthew Eappen. At the time, the medical community almost universally agreed on the symptoms of SBS. But starting around 1999, a fringe group of SBS skeptics began growing into a powerful reform movement. The Woodward case brought additional attention to the issue, inviting new research into the legitimacy of SBS. Today, as reflected in the Edmunds case, there are significant doubts about both the diagnosis of SBS and how it's being used in court.

In a compelling article published this month in the Washington University Law Review, DePaul University law professor Deborah Teurkheimer argues that the medical research has now shifted to the point where U.S. courts must conduct a major review of most SBS cases from the last 20 years. The problem, Teurkheimer explains, is that the presence of three symptoms in an infant victim—bleeding at the back of the eye, bleeding in the protective area of the brain, and brain swelling—have led doctors and child protective workers to immediately reach a conclusion of SBS. These symptoms have long been considered pathognomic, or exclusive, to SBS. As this line of thinking goes, if those three symptoms are present in the autopsy, then the child could only have been shaken to death.

Moreover, an SBS medical diagnosis has typically served as a legal diagnosis as well. Medical consensus previously held that these symptoms present immediately in the victim. Therefore, a diagnosis of SBS established cause of death (shaking), the identity of the killer (the person who was with the child when it died), and even the intent of the accused (the vigorous nature of the shaking established mens rea). Medical opinion was so uniform that the accused, like Edmunds, often didn't bother questioning the science. Instead, they'd often try to establish the possibility that someone else shook the child.

But now the consensus has shifted. Where the near-unanimous opinion once held that the SBS triad of symptoms could only result from a shaking with the force equivalent of a fall from a three-story to four-story window, or a car moving at 25 mph to 40 mph (depending on the source), research completed in 2003 using lifelike infant dolls suggested that vigorous human shaking produces bleeding similar to that of only a 2-foot to 3-foot fall. Furthermore, the shaking experiments failed to produce symptoms with the severity of those typically seen in SBS deaths.

The research implies that human beings simply cannot shake a baby to death without an accompanying impact to the head. SBS cases, however, frequently show no external injuries. This suggests that other causes are at work. Additional research has shown babies to be lucid up to 72 hours before classic SBS symptoms set in, casting doubt on the long-held theory that the child's caretaker at the time of death (or loss of consciousness) was the likely killer.

Last year, Discover magazine published a provocative article laying out much of this new research. Notably, the magazine found several specialists who have since changed their minds after testifying for the prosecution in multiple SBS cases. (At a post-conviction hearing for Edmunds, all of her defense experts said that when the case was tried in 1995, they would have testified for the prosecution.) One of those specialists is Ronald Uscinski, a student of Ayub Ommaya, the scientist whose research on monkeys in the late 1960s is thought to be the origin of the SBS diagnosis. When Uscinski went back and reexamined the study, he found no support for the way Ommaya’s research is currently being being used in the courtroom.

"When I put all of this together, I said, my God, this is a sham,” Uscinski told Discover. "Somebody made a mistake right at the very beginning, and look at what’s come out of it."

Teurkheimer estimates that "what's come out of it" is about 200 SBS prosecutions per year in America, mostly for murder. She believes there's legitimate reason to review nearly all of these cases, including even those where a suspect admitted to having shaken the baby. As she points out, suggestive or coercive questioning may have elicited such admissions. Moreover, in some cases, a defendant who admitted to shaking a child in order to revive it after it had already been unconscious was seen as having confessed to killing the child.

Whether someone can actually "free shake" an infant to death remains hotly disputed in the medical community. Where there is consensus, however, is that the triad of symptoms traditionally associated with SBS are not exclusive to it. A number of other things can produce these symptoms, including falls, head impacts, infections, birth defects, reaction to vaccinations, and surgical procedures. That's a significant departure from what prosecutors have been telling juries for the past 20 years.

In other words, there are almost certainly a significant number of innocent people in prison today who were wrongly convicted of shaking a baby to death. The problem is that there are also likely a number of guilty people who, nevertheless, shouldn't have been convicted on the basis of science-based testimony we now know to be false. The task will be convincing both the courts and the public to risk freeing actual child killers in order to free the innocent people convicted with flawed medical testimony.

Furthermore, unlike with DNA testing, which came about through rapid scientific breakthroughs, the issue of SBS is tied to a slow shift in the scientific consensus. We simply won’t have the slam-dunk evidence DNA provides when it points to the real culprit. With SBS, the question is usually whether a crime was even committed, or if a child's symptoms were caused by something other than shaking.

This whole controversy speaks to a fundamental tension between science and law. Science moves along a slow trajectory from inquiry toward certainty. While the courts have been eager to embrace new science—particularly forensic science—at the trial level, they're reluctant to revisit those cases when the science changes. One example is the now-discredited specialty of identifying bite mark evidence. But while science is mostly interested in testing, revising, and improving existing theories, once the jury has delivered its verdict, our criminal justice system puts a premium on finality. It takes a major upheaval in the scientific community (like DNA technology) to get courts to consider reopening old cases.

But at the very least the courts should stop prosecutors from making the same mistakes in the future. But even that isn't happening. Tuerkheimer, for example, found literature in current manuals for prosecutors that relies on discredited research from the 1980s and 1990s, still touting the pathognomic nature of SBS symptoms. And the same week Edmunds was given a new trial, an appeals court in Arkansas denied a new trial to a woman convicted under similar circumstances, based on the presence of the same symptoms.

Britain, Canada, and Australia have all initiated major reviews of shaken baby prosecutions in response to new research. Teurkheimer makes a convincing case that it’s time for the U.S. to do the same.

Radley Balko is a senior editor at Reason magazine.