Can Forensic Science Be Trusted?

Illustrations by Isabel Seliger

On February 12, 1981, 16-year-old Sherry Parsons returned home from high school in the small town of Norwalk, Ohio, and found a strangely quiet house. She called out for her mother, Barbara; hearing no response, she climbed the stairs and walked into her parents’ bedroom. “Then my eyes focused on the blood on the bed,” she recalled when I spoke with her recently. “I saw my mother on the floor, bludgeoned to death. I dropped my schoolbooks and started screaming.”

Blood soaked her mother’s nightgown and the bedsheets, and covered the walls and the ceiling. The police in Norwalk interviewed James Parsons, Barbara’s husband and Sherry’s father. There had been marital problems, but Parsons had a strong alibi: He had picked up breakfast at a coffee shop on the way to work at his auto-repair shop, where he saw customers throughout the morning. Police did not seriously investigate any other suspects.

The case was cold for about a decade, until Sergeant Mike White, in Norwalk, began looking into the murder. White wondered if he could connect the bedsheets to what he believed might have been the murder weapon: a Craftsman breaker bar—a heavy tool with a long handle, used to unscrew tight bolts—that had been found in a car that James Parsons had once owned. White approached the Cuyahoga County coroner’s office, in Cleveland. The technicians there examined the bedsheets and the tool, which had no traces of blood on it, and said they could not conclusively rule out the breaker bar as the murder weapon or connect it to the crime.

White then brought the matter to the Ohio Bureau of Criminal Investigation, where the case was assigned to a forensic scientist named G. Michele Yezzo, a bloodstain specialist. Yezzo proved to be more helpful. She believed she could make out a letter N, consistent with the appearance of the same letter in the word Craftsman on the breaker bar, imprinted on a bedsheet. She also believed that some stains on the victim’s nightgown—which are not easy to decipher—appeared to be similar in shape to the head area of the bar. She sprayed a chemical on the bedsheet and the nightgown to enhance the stains and raise any other impressions. As she watched, more detail emerged. She later testified that she was able to see “individualizing characteristics”—marks seemingly unique to that breaker bar—on the nightgown. She also testified that the letter S rose to the surface of the bedsheet—likewise consistent with the appearance of that letter in the word Craftsman. But Yezzo failed to photograph the newly visible image, and it faded. Moreover, the chemical process used to bring out the bloodstain markings—all of them, on both the bedsheet and the nightgown—made replication by the defense impossible. When asked, years later, why she had failed to photograph what she said she’d seen on the enhanced bedsheet, Yezzo replied, “This is one time that I didn’t manage to get it soon enough.” She added: “Operator error.”

In 1993, 12 years after the crime, James Parsons was indicted for the murder of his wife. The largely circumstantial case rested in no small part on G. Michele Yezzo—that is, on her credibility as an expert, including her unverifiable memory of what she may have seen when she conducted her experiment. Yezzo’s testimony provided a crucial physical link between Parsons and the crime. At trial, Yezzo acknowledged that other Craftsman tools—of which there are millions—were imprinted with the same logo. “I want to see more to be able to say it’s that bar, absolutely, to the exclusion of all others,” she said. But, she testified, “my opinion is that there is nothing that makes it inconsistent with this bar.”

If you are a semanticist, parsing carefully, those words mean little. In court, they can come across as definitive: Nothing rules out the possibility. The words were deployed as definitive by prosecutors—“the evidence is uncontroverted by the scientist, totally uncontroverted”—and understood that way by the jury. Parsons was found guilty and given a prison term of 15 years to life. Michael Donnelly, now a justice on the Ohio Supreme Court, did not preside over this case, but he has had ample exposure to the use of forensic evidence. “As a trial judge,” he told me, “I sat there for 14 years. And when forensics experts testified, the jury hung on their every word.”

illustration with fuzzy image of bedroom, microscopic splatter, illustration of head of tool
James Parsons was found guilty of murdering his wife based in part on bloodstain evidence that prosecutors said connected the crime to a Craftsman
tool. (Illustration by Isabel Seliger. Source: Sepia)

Two decades later, in 2013, the Ohio Innocence Project decided to look into the case. Parsons was still in prison. Because his conviction rested substantially on Yezzo’s testimony, the Innocence Project requested her personnel file from the Bureau of Criminal Investigation. “It was really just a Hail Mary,” Donald Caster, a professor at the University of Cincinnati College of Law and a staff attorney at the Ohio Innocence Project, told me recently. The legal team, he said, had noticed “squirrelly things” in a couple of other Yezzo cases. In early 2015, the Innocence Project received the bureau’s personnel file—all 449 pages of it. “People just don’t have personnel files that are hundreds of pages long,” Caster noted. “It’s not really a thing.”

The allegations in the personnel file detailed a long, acrimonious history. Among them: Yezzo had threatened to kill her co-workers, had threatened to kill herself, had threatened to bring a gun to work, had hurled a property-room key attached to a six-inch metal plate at a colleague, and had used a racial slur to describe a Black co-worker. With respect to her scientific analysis: “Her findings and conclusions regarding evidence may be suspect,” Daniel Chilton, the assistant superintendent of the Bureau of Criminal Investigation, had written in a memo to his boss in May 1989. “She will stretch the truth to satisfy a department.” Another note, written a few days before she was to testify in the Parsons trial, stated that Yezzo had a “reputation of giving dept. answer [it] wants if [you] stroke her.” In the same documents, analysts reworking some of Yezzo’s cases questioned her conclusions on a blood analysis and a partial-footprint analysis.

In an article about the personnel file that appeared in The Columbus Dispatch in 2016, Yezzo denied using a racial slur. She acknowledged problems with her behavior, citing personal pressures and overwork, but stated that her performance on the job had been unaffected.

[From the August 2019 issue: Barbara Bradley Hagerty on why police fail to catch many serial rapists]

The Bureau of Criminal Investigation documents revealed a suspicious timeline. Because of the allegations about her workplace behavior, Yezzo had been put on administrative leave two months before the Parsons trial. But her testimony was going to be needed. Yezzo was reinstated shortly before the trial. Parsons’s attorney was never told about the administrative leave or the bureau’s concern that her analyses might tilt in a single direction: toward law enforcement. “She should have never even testified,” Sherry Parsons told me. “But they got her to come back the day of the trial.”

In April 2016, after being presented with the new findings, an Ohio judge vacated Parsons’s conviction—not on the grounds that Yezzo’s analysis had been explicitly discredited but because the state had not revealed what it knew about possible bias. “This evidence could have been very useful to the defense in its cross-examination of Ms. Yezzo,” Judge Thomas Pokorny observed in an opinion from the bench. “Ultimately, these failures undermined his right to a fair trial.” He declared the guilty verdict “unworthy of confidence.” Parsons was released 23 years after his conviction, a fragile man suffering from heart disease, cancer, and dementia. He died 10 months later, at age 79. Sherry Parsons and her sister have now brought a civil suit against Yezzo, alleging that she violated James Parsons’s civil rights by fabricating evidence and withholding information about her administrative leave. (In a court filing in March, Yezzo denied these allegations. The case is ongoing.)

Forensic science, which drives the plots of movies and television shows, is accorded great respect by the public. And in the proper hands, it can provide persuasive insight. But in the wrong hands, it can trap innocent people in a vise of seeming inerrancy—and it has done so far too often. What’s more, although some forensic disciplines, such as DNA analysis, are reliable, others have been shown to have serious limitations.

A number of cases involving G. Michele Yezzo’s work in Ohio offer an entry point into the subject. I have read thousands of pages of court and other documents related to Yezzo’s career. Her alleged professional and personal shortcomings are a matter of record. That said, it is important to be clear about what she does and does not represent.

Yezzo is not like Annie Dookhan, a chemist in a Massachusetts crime laboratory who boosted her productivity by falsifying reports and by “dry labbing”—that is, reporting results without actually conducting any tests. At one trial, Dookhan testified that the substance a man had been caught with was crack cocaine when it in fact was a piece of peanut. The man served 15 months. Massachusetts has dismissed convictions in more than 20,000 cases Dookhan was involved in.

Nor is Yezzo like Michael West, a forensic odontologist who claimed that he could identify bite marks on a victim and then match those marks to a specific person. In their book, The Cadaver King and the Country Dentist, Radley Balko and Tucker Carrington recount how, in a videotaped autopsy of a 23-month-old girl, West was seen pressing a dental mold that he had made of a suspect into the toddler’s cheek, elbow, and arm. (West has called allegations that he was tampering with evidence “a damn lie.”)

The deeper issue with forensic science lies not in malfeasance or corruption—or utter incompetence—but in the gray area where Yezzo can be found. Her alleged personal problems are unusual: Only because of them did the details of her long career come to light. And yet the career itself is not as unusual as one might wish. It highlights how tenuous many forensic findings can be; how easy it is for prosecutors to make them appear solid to a jury; how closely some analysts work with law-enforcement colleagues, to the point of alignment; how rarely an analyst’s skills are called into question in court; and how seldom the performance of crime labs is subjected to any true oversight. All of this combines to create a dangerous prosecutorial weapon.

Brandon Garrett, a professor at Duke University’s law school and the author of Autopsy of a Crime Lab, told me recently that when he examined the forensic testimony in hundreds of wrongful convictions, he found “a blizzard of error.” To take a single metric: More than half of those exonerated by post-conviction DNA testing had been wrongly convicted based on flawed forensic evidence.

When asked to explain why forensics goes wrong, critics cite three factors. First, some commonly used forensic methods have not been rigorously evaluated; their validity has not been established. Second, the overwhelming majority of crime labs are not independent but tucked into police departments or state law-enforcement agencies. They depend on law enforcement for funding and operate under ever-present financial and psychological pressure to collaborate in securing convictions. Third, no one from the outside is rigorously checking the work done by forensic analysts, who may or may not have adequate scientific training. Many labs participate in voluntary accreditation programs, but, Garrett noted, accreditation largely focuses on having the right procedures spelled out on paper; proficiency tests given to lab analysts are extremely easy. Only in a few states are the regulations and controls that govern clinical labs and hospitals extended to crime labs as well. The quality of the work done in crime labs is almost never audited.

Even the best forensic scientists can fall prey to unintentional bias. A case widely cited by criminal-justice experts involved the FBI. In 2004, a series of explosions tore apart four trains in Madrid, killing nearly 200 people. The bombing was the work, it was believed, of Islamist terrorists. The Spanish National Police recovered a partial, smudged fingerprint on a bag of detonators and asked the FBI to analyze it. “Partials” are not ironclad and leave considerable room for interpretation. The FBI’s fingerprint database offered up 20 possible matches. An FBI examiner found minutiae in the fingerprint that led him to identify a U.S. citizen named Brandon Mayfield as the man who had touched the detonator bag. After reviewing the comparison, two other examiners agreed with this analysis. The FBI launched an investigation into Mayfield and learned that he was a lawyer in Oregon who had converted to Islam and once represented a Muslim man (in a child-custody case) who was later convicted of terrorism. Mayfield was soon in custody. The Spanish police never agreed with the FBI analysis, but their arguments initially went nowhere; ultimately the Spanish identified the print as belonging to an Algerian national. Two weeks after bringing Mayfield in, the FBI admitted its error and released him; the federal government later paid Mayfield $2 million in a settlement. A 2006 Inspector General report acknowledged that Mayfield’s background likely contributed to the FBI’s failure to reconsider its position after “legitimate questions” were raised.

Study after study has demonstrated the power of cognitive bias. Itiel Dror, a cognitive neuroscientist at University College London, found that forensic examiners analyzing a DNA mixture in a gang-rape case were far more likely to discern a possible link to a specific individual if they had already been told that the individual was a suspect. In a recent study, forensic anthropologists were also far more likely to wrongly conclude that a femur came from a female when they were shown staged images of a recovery scene containing a bra.

Cognitive bias can of course affect anyone, in any circumstance—but it is particularly dangerous in a criminal-justice system where forensic scientists have wide latitude as well as some incentive to support the views of prosecutors and the police.

Timothy Howard’s account of what happened on the morning of April 1, 2006, seemed straightforward. He had found his wife, Delilah, hanging from the belt of her pink terry-cloth bathrobe in the basement of their home in Franklin County, Ohio. Howard, a 44-year-old house painter, said that he’d cut Delilah down. He’d called 911. He’d performed CPR. But she was gone. After the police arrived, he’d gathered his three children—two daughters, 18 and 22, and a son, 11—and they’d settled in a bedroom. “One of the kids was looking on the dresser and found there was four notes, one to me and one to each of the three kids,” Howard told me recently at the public defender’s office in Columbus. “And they explained that she was sorry, but she can’t—couldn’t—go on.”

The marriage was strained, but Howard had no record of violence—indeed, no police record of any kind—and his wife had a history of depression. She had tried to commit suicide twice before. Nevertheless, Howard was charged with murder. Prosecutors alleged that Howard had strangled his wife with her bathrobe belt and staged the death as a suicide, advancing a bizarre theory that he had planted notes that Delilah had written in a previous suicide attempt.

Howard’s trial, in 2008, quickly turned into a battle of the experts. The state’s witness from the coroner’s office—not yet board-certified—concluded that Delilah had died from ligature strangulation. She ruled the death a homicide. Howard’s lawyers presented their own pathologist, who testified that the angle of the marks on her neck pointed to a classic hanging injury. The state presented an expert who insisted that Delilah weighed too much for a nail to hold. A defense expert disagreed.

illustration with fuzzy image of chair, nail, fiber, and belt
Timothy Howard said he found his wife, Delilah, hanging from the belt of her bathrobe in their basement. He was later convicted of murder in a case that involved fiber evidence. (Illustration by Isabel Seliger. Source: Sepia)

When G. Michele Yezzo, from the Bureau of Criminal Investigation, was called to the stand, she focused tightly on the fiber evidence. Yezzo explained that she had examined the “tape lifts” that had picked up debris on three nails near the spot where Howard said he’d found Delilah hanging. The tape lifts “failed to reveal any fiber samples that were consistent with the belt” from the robe. The prosecutor went over the point again, more bluntly this time, asking Yezzo to confirm that she had not found “any fibers.” She answered, “That is correct.” Howard’s defense relied in part on connecting the belt with a nail. Yezzo’s assessment was therefore crucial. “It’s easy to see how that testimony could have essentially been the tiebreaker,” Joanna Sanchez, the director of the Ohio public defender’s Wrongful Conviction Project, told me, “and pushed the jury to think, Okay, the physical evidence isn’t matching up with what he said.” Howard was convicted and given a prison term of 23 years to life.

In 2011, Sanchez began looking into Howard’s conviction. After agreeing to handle his case, she filed numerous public-records requests. Among the documents she received were Yezzo’s handwritten notes about her investigation. The notes indicated the presence of “no fine pink fibers”—as she had testified—but there were a few additional words: “some red noted but microscopically different in color.” Whatever the source and nature of the fibers, the fact of their existence had been omitted from Yezzo’s official report. She did not mention it in courtroom testimony when asked by the prosecutor about the presence of “any fibers,” thinking, she later explained, that the question was about the fibers she considered “consistent with” the belt. Sanchez believes that information about the presence of fibers would have been used to devastating effect by Howard’s defense team.

In 2012, a far more experienced coroner in Franklin County took a second look at the forensic findings, along with the health-care records that described Delilah’s previous suicide attempts. The coroner changed the manner of death from “homicide” to “undetermined.” With that, the state’s case was significantly undermined. Prosecutors fought for another five years, but in 2017, at age 56, Howard was released. While awaiting a hearing to be granted a new trial, he accepted what is known as an Alford plea: He was given his freedom in return for a guilty plea that included an assertion of his innocence.

In 2015, after James Parsons’s lawyers had requested Yezzo’s personnel files, Mike DeWine, at the time Ohio’s attorney general and now its governor, appointed a special prosecutor to investigate the forensic analyst’s performance. The audit largely focused on six years during which formal complaints had been made about Yezzo’s behavior, but even this limited review raised questions about her 32-year tenure. Of the 126 “death related” cases that involved Yezzo during this period, auditors flagged 18 for further review. In these instances, the audit took issue with Yezzo’s documentation—in some cases, for instance, because her notes were “minimal”; in some other cases because her findings did not support her final conclusions. Although Yezzo usually passed proficiency tests, in the 2000s, she failed tests involving hair and glass, and was reprimanded for a quality issue in a paint analysis; supervisors ordered dozens of her cases to be reexamined by other forensic scientists.

Yezzo had difficulty getting along with colleagues, but law-enforcement officers and prosecutors showered her with praise. “If you were here right now, I would give you a heartfelt hug and kiss for Valentine’s Day,” one prosecutor wrote. “There were several turning points in the trial. Your testimony was one of those critical times for us.” Another prosecutor called her “an important member of our law enforcement team.” A third referred to a successful prosecution as “a real collaborative venture.” This kind of reinforcement is not atypical. “Forensic experts get to know the prosecutors and they get to know the police,” Justice Donnelly, of the Ohio Supreme Court, observed during a recent conversation. “The lines can be blurred, and it begins to take on the semblance of a team effort. You’re all part of ‘Team Ohio.’ ”

[Listen: Barbara Bradley Hagerty talks about how innocence becomes irrelevant on the Radio Atlantic podcast ]

Yezzo maintains that she was never pro-prosecution; the audit put in motion by DeWine found “no indication of foul play or intentional deception” on Yezzo’s part and recommended that no further action be taken. But by that time Yezzo was long gone. She had been reprimanded by her superintendent in 2009 for “interpretational and observational errors” that “could lead to a substantial miscarriage of justice.” Yezzo resigned “in anticipation of retirement” a month later.

Prosecutors have noted that Yezzo’s testimony may not always have been pivotal. Ryan Stubenrauch, an attorney and former adviser to DeWine, told me that, with the exception of DNA evidence, “in your average case, it is not one piece of forensic evidence that makes or breaks” a prosecution. Witness testimony, motive, circumstantial evidence, other physical evidence—these combine to build to a conviction. Yet, as criminal law recognizes, one can’t simply remove a piece of load-bearing evidence after the fact and insist that the edifice would have remained standing anyway. Prosecutions are like a game of Jenga, Donald Caster, the Ohio Innocence Project attorney, observed. The government presents layers of evidence to create a tower of culpability. “If you pull the wrong piece out, the whole thing comes tumbling down,” he said. “And that’s what happens once you take Yezzo’s testimony out of a lot of these cases.”

Perhaps no single case demonstrates the outsize role that questionable forensic science can play in a defendant’s fate than the murder trial of Kevin Keith, a case that also illustrates the difficulty of true legal redress.

On February 13, 1994, at about 8:45 p.m., a man came to the door of an apartment in the small city of Bucyrus, Ohio, about an hour north of Columbus. The man spoke briefly with a woman named Linda Chatman, who had just arrived; the apartment belonged to Marichell Chatman, Linda’s niece. Whoever the man was, he was known to Linda and Marichell. He walked inside, asked who was winning the basketball game on television, then pulled a 9-mm handgun and ordered all six people there—three adults, three children—to lie on the ground. “What are you doing?” Marichell asked, according to later trial testimony. “You ain’t going to hurt us while the children are here. I don’t want anything to happen to those children.” “You should have thought about this before your brother started ratting on people,” the man replied, and shot Marichell in the head. Then he trained his gun on the others. When he was done, Linda, Marichell, and Marichell’s young daughter lay dead. Marichell’s boyfriend and two young cousins were badly wounded but would survive. Glancing out her front door, a resident of the complex saw a husky Black man careening out of the parking lot in a cream-colored automobile. It skidded into a snowbank and eventually sped off after the driver got out of the car and rocked it free.

Police immediately suspected a man named Kevin Keith. Keith was a small-time drug dealer who had recently been charged with drug trafficking, thanks to evidence supplied by Marichell’s brother Rudel Chatman. Police arrested Keith two days after the crime as he sat on his couch watching TV. When I visited him last year at the medium-security prison in Marion, Ohio, he told me he’d thought they were picking him up for selling “a little weed.” He went on: “The next thing I know, I’m being read murder charges.”

At Keith’s trial, in 1994, the prosecutor promised a case supported by numerous eyewitness accounts and compelling forensic evidence. Jurors got something else. One of the young cousins who survived the attack, a 6-year-old girl, had told police explicitly that Keith was not the culprit. “It looks like him but that’s not him,” she said. When it came to identifying the perpetrator, witnesses contradicted one another and themselves, and even disagreed about whether or not he had been masked. No physical or biological evidence—blood, hair, fibers, fingerprints—connected Keith to the crime scene. Could Keith at least be connected to the snowbank? The getaway car had left tire-tread impressions and a partial license-plate impression—the numbers 043.

When Keith’s girlfriend came to visit him in jail, the police took notice of the license plate on the car she was driving: MVR043. However, the car was not cream-colored—it was green. And it was not Keith’s car or even his girlfriend’s car—it was her grandfather’s car. Moreover, the tires on the car did not match the tread marks in the snow. The only possibly good news for the prosecution was that the tracks in the snow seemed to resemble those from a type of tire, Triumph 2000s, that the grandfather had once owned—the police had found a receipt. The state’s theory seemed to be that Keith had replaced the Triumph 2000s with some other kind of tire after the crime. Or maybe that he had put them on to commit the crime and then changed them back. “Why would somebody change the tires on a car and not the license plate?” Rachel Troutman, an attorney at the Ohio public defender’s office, asked when I spoke with her recently.

To fortify its case, the prosecution relied on Yezzo, who testified by deposition. She had herself conducted some of the analyses that found no physical link between Keith and the murder, but a Bucyrus police captain had sent her a brochure advertising Triumph 2000 tires along with a copy of the grandfather’s receipt, and had scrawled a note: “Hope this will do the trick for us.” Yezzo never looked at the actual tires that the state alleged had made the impression in the snowbank—the ones that the grandfather had once bought but that were not on the car when the police examined it. Rather, she simply compared photographs and a plaster cast of the tire-tread impressions with images from the brochure. She testified that the tracks in the snow were similar to the tread design on the type of tire that had once been on the grandfather’s car: “Its similarity is it would have originated from the Triumph 2000,” she stated. (Yezzo has taken issue with the court transcript, believing that she would have said “could have.”) As for the number sequence on the license plate, Yezzo reported that the impression in the snowbank had “spacing and orientation similar to the license plate ‘MVR043,’ ” meaning that, in her judgment, the numbers came at the far right of the license plate—a conclusion unfriendly to Keith and one that effectively ruled out anyone whose license plate had 043 on the far left. Even so, there were 17 other cars in tiny Bucyrus, and a total of 130 in Crawford County and nearby Richland County, that had the sequence 043 at the far right of their license plate.

illustration with fuzzy image of car in snow, license plate
In the murder trial of Kevin Keith, the state alleged that partial tire and license-plate impressions in the snow tied the defendant to the crime scene. (Illustration by Isabel Seliger. Source: Sepia)

Looking back at Yezzo’s testimony, Justin Herdman, a former U.S. attorney for the Northern District of Ohio, told me: “Comparing photographs is about the worst possible way you could ever think of to make a scientific determination about commonalities or draw connections. That’s not science. That’s somebody on the back of a cocktail napkin drawing conclusions.” Nonetheless, Keith was convicted and sentenced to death.

When Rachel Troutman first met Keith, in 2007, she was skeptical of his claim of innocence. She would soon change her mind. She discovered radio logs and police reports that called into question the veracity of witnesses and police officers. She found that the Bucyrus police had received two subpoenas for phone logs and other records that could have been exculpatory; someone had scrawled “Ignore” across the subpoenas. As outlined in a motion for a new trial, she also found that police had suppressed information from an informant about a man who allegedly told a friend he’d been paid to “cripple” Rudel Chatman, Marichell’s brother, because he was a snitch. This man, who had served time for murder when he was a minor, owned a light-colored car and had a license plate containing the sequence 043, though the numbers did not occur at the end, where Yezzo had said they ought to be.

Every time Troutman discovered new exculpatory evidence, the court declined to consider it. As Keith’s execution date drew near, in 2010, the parole board unanimously rejected a bid for clemency. Troutman finally appealed to Governor Ted Strickland. Thirteen days before Keith was to be executed, Strickland commuted his sentence to life without parole, noting that he was troubled by “important questions” about the forensic evidence and the failure to fully investigate “other credible suspects.”

[From the January/February 2018 issue: Can you prove your innocence without DNA?]

Prominent members of the Ohio bar have moved into Kevin Keith’s camp. Justin Herdman is now a member of his legal team. Lee Fisher, Ohio’s attorney general when Keith was tried, filed an affidavit in court on Keith’s behalf: “I am deeply concerned that Ms. Yezzo’s conclusions and testimony led to a miscarriage of justice in Mr. Keith’s case,” he wrote. In another affidavit, Strickland stated, “I have become convinced that Kevin Keith is very likely an innocent man.”

G. Michele Yezzo agreed to speak with me last year. We met at a deli, in a strip mall in Columbus. She is in her late 60s, with brown hair and bangs cut straight across her forehead. She wore a rainbow-colored T-shirt in support of efforts to combat autism. Although Yezzo would not talk about any of the allegations that appear in her personnel file, she insisted that she had never skewed her findings. She also said that she had sometimes testified for the defense. If a prosecutor uses her words a certain way, Yezzo maintained, she should not be held responsible. “I do the analysis to the best of my ability, and if anything, I’m conservative.”

Yezzo is already being sued for her role in the Parsons conviction, and other lawsuits may be in the offing. The Ohio attorney general’s office declined to comment about Yezzo, citing the ongoing litigation, and added that since Yezzo’s retirement, the Bureau of Criminal Investigation has developed “several quality controls.” When I mentioned her name to John Lenhart, who ran the bureau back in the early ’90s, he groaned. Lenhart had put Yezzo on administrative leave in 1993, shortly before the Parsons trial; she was reinstated without his knowledge, and kept her job for another 16 years, giving testimony in hundreds of criminal cases. Lenhart left the bureau in 1994 and returned to state government years later, as Ohio’s director of law enforcement. He was surprised to be told that Yezzo was still there. “I’m going, ‘You’ve got to be kidding me,’ ” he recalled.

When I spoke with Yezzo, she sought to diminish her influence over any particular outcome. “I am not the one to say, ‘That person did it,’ ” she explained. “I can say: ‘This is what I have from the physical evidence.’ That being the case, I’m not judge and jury, and never will be, and never have been.” None of this changes the fact that Yezzo’s judgments were given an imprimatur of expertise and certainty by prosecutors, or that juries did convict in part because of them.

Questions of forensic judgment aside, jurors and the public do not appreciate the fact that the “science” of forensics may not warrant the label. In 2009, in a comprehensive report called “Strengthening Forensic Science in the United States,” the National Academy of Sciences delivered a withering assessment. Under the academy’s auspices, experts of all kinds had been brought together to examine disciplines including the analysis of fingerprints, firearms, bite marks, and bloodstains. They drilled down on the question How good is forensic science at connecting a piece of evidence to a specific individual or source? The answer: Only DNA could reliably do this. Other disciplines had not yet been backed up by robust research.

Consider bite marks. For years, the American Board of Forensic Odontology claimed that it was possible to match a suspect’s teeth to the bite marks on a victim’s body with “virtual certainty,” despite the fact that skin is elastic and bite marks change over time. Some studies have found that forensic dentists match bite marks to the wrong person about 15 percent of the time. (Dental experts sometimes can’t agree whether a mark came from a human being.) Bloodstain analysts, who look at patterns of blood to re-create the backstory of physical violence—an assault by, say, a six-foot-tall, right-handed man with a hammer, standing over the victim—have fared little better than the dentists. In a recent study, forensic experts were shown photographs of a bloody scene for which the study’s authors knew the “ground truth”—that is, how the patterns had been created. The experts got it wrong about 11 percent of the time. Perhaps the most alarming finding involves comparative hair analysis using microscopes. This technique has now been largely superseded by DNA analysis, but it helped put untold numbers of people in prison and dozens on death row. By 2015, the FBI had reviewed hundreds of criminal cases in which FBI examiners had testified that hair found at a crime scene incriminated a suspect. The FBI concluded that its experts had provided scientifically invalid testimony in 96 percent of the cases—including 33 of the 35 death-penalty cases. “It’s hard to be that wrong,” Mark Godsey, the head of the Ohio Innocence Project, told me.

That doesn’t mean that “all forensic science is worthless,” Sandra Guerra Thompson, the author of Cops in Lab Coats, explained when I spoke with her. “Many disciplines are solid, but we just don’t have enough proof yet to fully support others.” She pointed out that serology, toxicology, chemical analysis, and of course DNA analysis rest on secure foundations. The larger problem, experts say, involves pattern-matching disciplines, which still overwhelmingly lack established methods or standards. They also require judgment calls—essentially, comparing two pieces of evidence and determining whether, for instance, a smudged, partial tread mark at a crime scene might have been made by a suspect’s shoe, or whether a fiber found in a suspect’s car could reasonably have come from a victim’s sweater. These comparisons may be more prone to bias or error.

The 2009 National Academy of Sciences report was a bombshell, at least among forensic professionals. But it had little impact, and proposed reforms, such as removing crime laboratories from state and local law-enforcement agencies and making them independent, rarely gained traction. When a presidential advisory council reported in 2016 that many forensic disciplines did not yet pass scientific muster, the Justice Department essentially rejected the report. The National Commission on Forensic Science, created in 2013 to raise the standards of the science used in courts, was disbanded during the Trump administration.

The story of G. Michele Yezzo reveals not that Ohio’s crime lab is uniquely bad—it is highly regarded, in fact—but that it is not unique at all. “There are so many different ways that forensics can go wrong,” Duke University’s Brandon Garrett told me. It’s not just about unreliable forensic methods and it’s not just about biased experts: “It’s about the whole forensics enterprise and how poorly we regulate it.” Garrett keeps a running tab of scandals at crime laboratories. Hardly a month goes by, he said, when he doesn’t find another lab to add to the list, whether for specific cities (Cleveland, New York, San Francisco, Detroit) or for entire states (West Virginia, Montana). Some crime labs have been closed; others, including in Washington, D.C., have lost their accreditation and have had to outsource their work.

When properly conducted and modestly characterized, forensic science has a clear role in the criminal-justice system. And a few labs have made changes that might preserve its utility while mitigating its problems. One bright spot is Houston. In 2003, after it became clear that faulty forensic work by Houston’s police crime lab would lead to an exoneration, The New York Times described the lab as “the worst” in the country. Auditors discovered that technicians were poorly trained, kept shoddy records, misinterpreted data, and submitted reports based on evidence that they hadn’t bothered to test—this in a state that puts more people to death than any other. “Houston was a hot, stinking mess,” Peter Stout, who took over as president of the laboratory in 2017, told me. “It got bad enough that everybody said, ‘Okay, we’ve got to do something different.’ ”

The old crime lab was scrapped. Taking its place was the Houston Forensic Science Center. No longer part of the police department, the center is a “local government corporation,” Stout told me, with a board of directors that includes defense attorneys and even one exoneree. The Forensic Science Center analyzes evidence for both the defense and the prosecution. It steers clear of dubious techniques, such as hair analysis, bite-mark comparisons, and bloodstain analysis. Crucially, it employs blind testing: Periodically, the lab’s managers slip a sample into the stream of evidence for which they know the ground truth in advance—this sample is cocaine; the owner of this latent print is in the database; this DNA mixture involves these three people. With such exercises, Houston can determine whether analysts are arriving at the correct answers. These kinds of procedures will not prevent every mistake. But they do introduce an element of forensic neutrality—and, over time, will perhaps lead jurors to a more realistic view of what weight to place on forensic analysis.

Meanwhile, only seven states have laws that explicitly allow a prisoner to challenge a conviction on grounds of discredited forensic science. Ohio is not one of them, and the path to a new trial in states without such laws can be difficult. As for attempting to bring a case in federal court, that too is often a dead end. In 1996, at the urging of President Bill Clinton, Congress passed a law that makes it nearly impossible for federal courts to overturn state convictions. For Kevin Keith, this means that even though his lawyers have discovered new information, he almost certainly cannot get relief. In 2021, after considering Yezzo’s testimony, the wealth of new evidence, and the applicable law, U.S. District Court Judge Solomon Oliver Jr. denied Keith a new day in court. He had no jurisdiction. “This case, more than most, demonstrates the tragic result” of the 1996 law, Oliver wrote. But his hands were tied. He had to apply the law as it is, “not as the court wishes it to be.” Kevin Keith remains in prison.

This article appears in the June 2022 print edition with the headline “Can Forensic Science Be Trusted?” When you buy a book using a link on this page, we receive a commission. Thank you for supporting The Atlantic.