Search About Donate
Filed 7:15 a.m.
06.24.2015

The Surprisingly Imperfect Science of DNA Testing

How a proven tool may be anything but.

Illustration by Andrew Dubbin/Fusion

Years later, none of it — not 84-year-old Eleonora Knoernschild’s bloody body on the shag carpet, not the torn bedspread twisted around her neck, not the junk heaped on her corpse so abundantly that only her left foot poked out, not three decades of detective work — none of it would matter as much as the cheese wrapper.

This story was produced in partnership with Fusion and FRONTLINE.

The day after Knoernschild was killed on Nov. 4, 1984, the local newspapers didn’t mention the cheese wrapper at all, nor the knee-high stocking that would also be of great consequence in the trials that would take place 30 years later.

Instead, the newspapers reported how Knoernschild’s premature death had been discovered: Her daughter, 59-year-old Doris Wines, had been walking to the neighborhood donut shop that Sunday morning. Wines lived just a few doors down from her mother in St. Charles, Mo., an affluent St. Louis suburb on the western bank of the Missouri River. She stooped to pick up the newspaper on her mother’s lawn. As she went to drop it off, she saw that the window in the front door had been broken. She sprinted home and called 911.

That afternoon, police told reporters that Knoernschild had been murdered and her home ransacked. The motive was likely burglary, they said, but so far there were no witnesses or suspects.

Detective Mike Harvey of the St. Charles Police Department was assigned to the case. Harvey was a Vietnam vet with a flair for convincing criminals to rat each other out. He would devote considerable reflection over the next 30 years to the death of Eleonora Knoernschild. Long after it waned in departmental memory, he was still following leads. Even after he retired in 2009, Harvey thought about her death. So when an old friend in the county prosecutor’s office called and asked if he’d like to come out of retirement to work on cold cases for them, he did not hesitate. The Knoernschild files were the first he opened.

That was in April 2010, when the cheese wrapper had not yet gained its prominence in the case, although it had been in the custody of law enforcement for many years. Detectives found the rectangle of translucent plastic on Knoernschild’s linoleum floor next to a frozen Yankee pot roast. Over the next 27 years, the cheese wrapper sat inside a manila envelope in the St. Charles Police Department’s property room.

In November 2010, Harvey asked the county forensics lab to test it and 14 other items from the crime scene for DNA. One day the following summer, DNA technician Dan Fahnestock slid the cheese wrapper from its envelope. Fahnestock, half of the county’s two-person biological forensics lab, examined the wrapper. It was covered in silver dust, a relic of fingerprint testing it had been submitted to years before.

He wiped the wrapper with six cotton swabs and placed them each in solution to dissolve any human cells they had collected. He ran the solution through a series of machines tasked with detecting, isolating, and amplifying DNA. For each of the samples, the robots produced an electropherogram, a chart with a series of flat, squiggly lines punctuated by spikes, like the EKG of a failing heart or a seismograph recording tiny earthquakes. Several of the swabs returned nothing useful, but one produced a chart with a handful of peaks, each representing a genetic marker.

Fahnestock examined the chart. It wasn’t ideal. The amount of genetic material wasn’t exactly abundant, and the DNA had decayed considerably since 1984. Some markers couldn’t be seen at all.

Conservative technicians might have stopped there, deciding the sample was too compromised to analyze. But Fahnestock continued, piecing together a profile. That profile pointed to someone Harvey had suspected for years — St. Charles native son, Brian McBenge, an ex-boyfriend of Knoernschild’s granddaughter. Six weeks later, a DNA test of a knee-high nylon stocking found behind Knoernschild’s house implicated his younger brother, Cecil McBenge.

The DNA evidence would serve as the basis for prosecuting the brothers for first-degree murder. It was a crime Harvey believed they had gotten away with for decades. But the cheese wrapper may be evidence of something else entirely: the ambiguity of a science most people never doubt.

Eleonora Knoernschild on her porch in St. Charles, Mo. in 1982.

T he man Harvey says killed Eleonora Knoernschild strode to the sturdy plastic table in the visiting room where I was seated. His goatee was sandy grey, same as his hair.

He placed a stack of papers between us: The 296-page deposition of Dan Fanhestock. In preparation for our meeting, Brian had reread the deposition, and on three sheets of yellow legal paper, in tidy print, listed dozens of page numbers where he spotted holes in Fahnestock’s DNA analyses of the cheese wrapper and stocking. He started to hand me the list, but a guard stopped him. Prisoners are forbidden from passing papers to visitors.

Brian sighed and forged ahead, flipping through the pages. He pointed to a question his attorney had asked Fahnestock: Whether Fahnestock could say with certainty that Cecil or Brian were the individuals who left the DNA on the cheese wrapper or the knee-high stocking.

Brian looked at me. “And he says no.”

Fahnestock’s answer was a simple acknowledgement that even at its best, DNA science is not absolute. To Brian, it was proof that DNA cannot be trusted.

In the three decades since DNA emerged as a forensic tool, courts have rarely been skeptical about its power. When the technique of identifying people by their genes was invented, it seemed like just the thing the justice system had always been waiting for: Bare, scientific fact that could circumvent the problems of human perception, motivation, and bias.

Other forensic sciences had taken a stab at this task. Lie detector tests, ballistics, fingerprinting, arson analysis, hair examinations — all aim to provide evidence independent of the flawed humans wrapped up in an investigation. But those methods were invented by law-enforcement agencies eager for clues; it is now well established that their results are not always sound. With few alternatives, police and courts spent most of the 20th century hammering away at justice with the rubber tools of traditional forensics.

DNA was different. It came up through science, which began, in the 1950s, to unravel the ways the double helix drafts our existence. When DNA profiling led to its first conviction in a U.S. courtroom in 1987, DNA had already vaulted through the validating hoops of the scientific method. Soon it was accompanied by odds with enough zeros in front of the decimal to eliminate reasonable doubt.

Today, most of us see DNA evidence as terrifically persuasive: A 2005 Gallup poll found that 85 percent of Americans considered DNA to be either very or completely reliable. Studies by researchers at the University of Nevada, Yale, and Claremont McKenna College found that jurors rated DNA evidence 95 percent accurate and between 90 and 94 percent persuasive, depending on where the DNA was found. That faith could be shaken, but only when lawyers made a convincing case that a lab had a history of errors.

Otherwise, the mere introduction of DNA in a courtroom seemed to stymie any defense. “A mystical aura of definitiveness often surrounds the value of DNA evidence,” the studies’ authors wrote.

In many cases, this aura is deserved. The method is unequivocal when it tests a large quantity of one person’s well-preserved genes, when it’s clear how that evidence arrived at a crime scene, and when the lab makes no errors in its work.

But those are not circumstances enjoyed by every criminal investigation. Take the case of Kerry Robinson of Georgia. Robinson was implicated, in part, when two analysts concluded his genes may be present on the victim’s vaginal swabs. The jury convicted, and Robinson received a 20-year sentence.

Greg Hampikian, a biology and criminal justice professor at Boise State University and director of the Idaho Innocence Project, was a defense expert in the trial and felt sure the analysts had reached their conclusion because of unconscious bias: They knew a great deal about the case, including that the detectives believed Robinson was guilty. To test his suspicions, Hampikian and cognitive neuroscientist Itiel Dror of University College London sent the DNA data to 17 other analysts and asked them to interpret it without any information about the case. Only one agreed with the original analysts.

Despite these results, the Georgia appeals court declined to overturn the conviction, stating that “as long as there is some competent evidence, even though contradicted … we must uphold the jury’s verdict.”

Because DNA is more reliable than other forensics, scientists have shrugged off suggestions that it could fall victim to the vagaries of bias. But Dror noted that much DNA analysis involves interpretation. With interpretation comes subjectivity, and with subjectivity can come error.

“DNA results can be in the eye of the beholder,” Dror said.

P olice officer John Young was the first to arrive at Knoernschild’s house. His partner conducted a perimeter check while Young stepped over the splintered glass on the porch. The door was locked, so Young reached through its broken window and opened it. He stepped inside and found himself in the dining room.

Two cabinets and a desk had been emptied; papers and pill bottles littered the floor. Kitchen cupboards were open and food cast around. Young approached a bedroom door, where a sign warned against smoking because an oxygen tank was in use. To the right, a bare mattress. To the left, a dresser emptied of its contents. On the floor, a heap of clothes and paper. He stared at it. Knoersnchild’s foot was sticking out.

He dug through the debris. Under a blue gown, he found her face. Blood ran out of her mouth and down her right cheek, forming a pool. A tube tethered her face to a nearby oxygen tank. The fringe of her bedspread encircled her neck. The coroner would conclude that blunt force trauma had killed her.

Crime-scene investigator Bob Brockmeyer arrived minutes later. He began directing a team that searched the house and neighborhood for clues. The Yankee pot roast was collected, as was the cheese wrapper. Two mismatched gloves were found in a hedge across the street; their twins were in Knoernschild’s bedroom. A knee-high stocking was discovered near the garage, and another in the alley behind the neighbor’s house. One of the footprints on the front porch had a funny mesh pattern; Brockmeyer hypothesized the perpetrator pulled the stocking over his boots so as not to leave footprints. These items and dozens more were placed in manila envelopes and marked as evidence.

Detective Harvey was assigned to the case that day. Over the next months, he chased leads, but nothing materialized. He would arrive at his theory that Brian had killed Knoernschild the following year, in part after realizing that Brian had once dated Knoernschild’s granddaughter, Debbie Wines.

An evidence photo of the food discovered on the floor of Knoernschild's kitchen. At top, circled in red, is the clear cheese wrapper.

In January 1986, Harvey interviewed Wines about her relationship with Brian. Wines had suffered a tumultuous adolescence, experimenting with booze, drugs, and boys by middle school. Her parents sent her to a series of boarding schools; she’d met Brian through friends at one of them. Soon, they were dating. She introduced him to her parents, and they thought he was well mannered but didn’t like him. Something felt phony about his politeness, Don Wines would later say. Sometimes Debbie and Brian would go over to Knoernschild’s house and she would sneak $10 or $20 out of the Calumet baking powder can where her grandmother kept cash. After a while, she heard that Brian had gone out with another girl and broke up with him.

Harvey asked Wines if she was still dating Brian in early 1980, when she was 15 and he was 18, and her grandmother’s house was burglarized the first time: Cupboards were opened, junk dumped onto the floor — an uncanny preview of the crime scene four years later. Police hadn’t solved the 1980 burglary at the time, but a review of case records revealed that a fingerprint consistent with Brian’s had been lifted from the refrigerator. Wines said she couldn’t recall exactly when they broke up, so she couldn’t say if his fingerprint might have arrived there innocently. Her parents had no better recollection.

Brian’s adolescence and young adulthood were peppered with petty theft. One day, he and a friend got drunk and stole a car. Highway patrol pulled them over, and they escaped by foot. They wound up at an uncle’s house in central Missouri and continued their spree. His uncle got home and realized his truck and boots were missing, and called the police. Brian and a friend were arrested.

Like his brother, Cecil had been in and out of jail for petty crimes as a youth. One day when he was 21, he dropped some friends off at a restaurant, knowing they intended to rob it. The three were caught and arrested; Cecil was convicted of attempted armed robbery and armed criminal action.

While Cecil was awaiting sentencing, Harvey came to visit him and asked what he knew about his brother’s crimes. Cecil told him that once, when his brother called from jail, Cecil asked him if he’d killed an acquaintance of theirs whose death Harvey was also investigating. Brian responded, “Is your birthday in January?” Cecil said yes. Brian replied, “Well, that’s your answer.” According to Cecil, he made up the story on a stupid, youthful whim in hopes of escaping a long sentence for the robbery charges. Harvey asked Cecil to take a lie detector test implicating his brother, but the results were inconclusive. He was sentenced to 30 years.

When Brian left prison in 1987, he married and moved to a small town in southeastern Missouri, where his wife ran a restaurant; Brian was her breakfast cook when he wasn’t working construction. They raised two daughters. In 2000, Cecil was released on parole 15 years into his 30-year sentence. The following year, he was caught with cocaine at a Sammy Hagar concert and sent back. He was released again in 2003 and fell in love with an accountant named Sue. They had a daughter, adopted a dog, and bought a house.

During those years, Harvey continued investigating the brothers. He became increasingly convinced that Brian was not only responsible for Knoernschild’s death, but for two other unsolved murders from the 1980s. After joining the prosecutor’s office, he redoubled his investigative efforts, and in 2012, Brian was arrested on homicide charges in the two other cases. Harvey also led an investigation into allegations that Cecil had stabbed a fellow inmate in the early 1990s.

To Harvey’s disappointment, those cases proved to be dead ends: Brian was acquitted by a jury after 28 minutes of deliberation, and the other homicide case may not go to trial because the sole witness is now facing his own legal troubles. A judge could not find evidence that Cecil was guilty of the stabbing, and those charges were dropped. The first-degree murder charges for the death of Knoernschild, however, would prove more problematic for the McBenges.

I n photos taken the morning of the murder, a police officer grasps Knoernschild’s thigh and arm, lifting her limp body to expose the blood on her bedroom carpet. In the picture, the officer is barehanded.

People familiar with his work from the 1980s say that Brockmeyer, the crime-scene investigator, and his team probably weren’t wearing masks the day they searched Knoernschild’s house for clues, and likely used brushes and equipment that was not often cleaned between crime scenes (Brockmeyer died of a heart attack in 2003). The following day, Brockmeyer checked the stocking into the St. Charles Police Department’s property room, but he waited to turn over the cheese wrapper and several other items for nine days. He may have held onto them to do his own fingerprint processing, but it’s unclear — there is no record of the cheese wrapper’s whereabouts during that time.

The stocking was eventually sent to Southeast Missouri University (SEMO) in Cape Girardeau, where Professor Robert Briner trained graduate students in forensic analysis in a house-turned-lab. According to his records, he tested the stocking for saliva secretions, and found none. Enough time has passed that he does not remember specifically if he used a mask, or what evidence from other cases were tested in the same workspace, or which graduate students may have been around during testing.

Several months later, the stocking was sent back to St. Charles, where it joined the cheese wrapper in the evidence room. They sat there for more than two decades, until they were sent to Fahnestock for DNA testing. The cheese wrapper arrived at the crime lab in a sealed envelope; the stocking in an unsealed one.

As Fahnestock handled the evidence, he wore two layers of gloves, a facemask, and the evidence only touched surfaces that were covered with clean sheets of butcher paper. Despite these precautions, Fahnestock found his DNA on an item from Knoernschild’s house: A sample he tested from a blue glove discovered at the crime scene included at least two DNA profiles, one of which appeared to be his own.

The cheese wrapper prior to being tested for DNA in 2011.

Scientists are still exploring the circumstances and ease with which DNA can travel. Many of our cells and fluids — skin, saliva, sweat, and mucus — routinely find their way into our environment. If conditions are favorable, our genes can wind up places we’ve never been. After Silicon Valley millionaire Raveesh Kumra was killed in his 7,000-foot mansion in November 2012, police discovered the DNA of Lukis Anderson, a 26-year-old homeless man, on his fingernails. But hospital records indicated that Anderson was unconscious in a hospital bed while Kumra asphyxiated nine miles away.

Anderson spent five months in jail while lawyers and investigators pondered how he could have committed the crime. Finally, they realized that the paramedics who transported Anderson to the hospital had also responded to the homicide. They had clipped an oxygen-monitoring probe to Anderson’s finger that morning, and to Kumra’s that afternoon. Anderson’s DNA had gone along for the ride.

“It's a small world," Santa Clara County Deputy District Attorney Kevin Smith told the San Francisco Chronicle after the mistake was discovered.

Peter Gill is a giant in the forensic DNA community, counted among the scientists who wrote the original paper conceptualizing DNA as a forensic tool in 1985. But he has spent recent years warning people using his tool against blindly trusting its results. In a 2014 book called “Misleading DNA Evidence: Reasons for Miscarriages of Justice,” Gill wrote that contamination is dangerous because investigators are eager to believe that DNA found at a crime scene must come from the perpetrator.

“The presence of a DNA profile says nothing about the time frame or the circumstances under which it came to be there,” says defense expert and researcher Dan Krane. “Test results can’t distinguish between the possibility of contamination, or evidence tampering, or, you know, murder.”

Dan Fahnestock, DNA technical leader for the St. Charles Sheriff's Department crime lab in June.

Technology may soon increase the danger of implicating innocent people. Today, most DNA analytical machines are optimized to parse the DNA of about 100 human cells. Future generations of forensic robots may extract a profile from just one. The DNA of a person who drives by a crime scene with an open window could wind up somewhere suspicious; shake someone’s hand before he commits a crime, and you may be implicated.

“Next gen sequencing might generate lots of data from small amounts of DNA, but you’re still faced with the same fundamental question — what does it mean?” said John Butler, special assistant to the director for forensic science at the National Institute of Standards and Technology, who has written several textbooks on DNA analysis. “You could detect a single cell on a knife blade, but that doesn’t mean anything — it might have arrived there long before the crime or been transferred there by chance.”

Everyone on the McBenge brothers’ defense team has a theory about how Cecil and Brian’s DNA could have found its way onto the evidence; some are more far-fetched than others. Cecil’s attorney Cyndy Short wondered if Knoernschild had tossed the stockings out by the trash cans days before she died, and Cecil, who lived a mile away, walked down the alley and sneezed. Bicka Barlow, another attorney and DNA expert on Cecil’s defense team, noted that a refrigerator repairman had visited the house earlier that week — perhaps a cheese wrapper that Brian had touched when he was dating Debbie had been sitting behind the fridge for years before it was somehow loosened onto the crime-scene floor. DNA expert Dan Krane of Wright State University in Dayton, Ohio, who testified for the defense, speculated that the investigators’ fingerprint brushes could have transferred Brian’s DNA onto the wrapper. “It doesn’t matter how long before the crime [Brian] was in the house, all that matters is that he was,” Krane said.

Eighteen pieces of evidence were tested from the crime scene, including knives found in the bedroom, the ligature around her neck, and the nightgown she was wearing while she was killed. Had the McBenges’ DNA appeared on anything intimate to the crime, the defense believes there would be less to speculate about. “But it was on a damn cheese wrapper,” said Short. And a cheese wrapper, she said, doesn’t tell the story of a murder.

The nylon stocking found in the back alley was even farther from the crime scene. Investigators have toyed with how it may have been deployed: First, they theorized it was pulled over a boot. Later, they speculated it was worn as a glove or a mask.

Prosecutor Phil Groenweghe said he doesn’t need to know exactly how it was used. “There are limits on what we can say happened because the only eyewitness is dead,” he said. “In a criminal prosecution, there’s certain elements that have to be proven, and the point in time when they put on the nylons isn’t one of them.”

Groenweghe disputed that the DNA could have arrived on the cheese wrapper or stocking innocently. “DNA was found on items that clearly had been touched at the time of the crime. We didn’t find it on the door jam.”

He also dismissed any theory that placed the DNA on the two items by accident. “Unless Brian and Cecil were working as lab techs at SEMO, I don’t think that’s a very compelling defense,” he told me. “The fact is, it was their DNA on the evidence.”

But the McBenges challenge whether it was their DNA at all.

I n the three years between the McBenges’ arrest and their trial, Fahnestock worked hard to make sure that the DNA he’d found didn’t belong to any of the bare-handed people who had handled the evidence. He tested the DNA samples of half a dozen current and retired police officers. He tested Dr. Robert Briner of SEMO. He tested everyone in Knoernschild’s family. He tested Debbie’s live-in boyfriend from 1984. He tested the cleaning lady who had waxed Knoernschild’s floors. He even tested Brockmeyer, who had passed away years earlier, by testing his son and wife.

As Fahnestock has explained to dozens of juries, crime labs don’t exactly map the human genome. Instead, they typically focus on 13 places, or loci, plus a 14th that expresses gender. Each loci is home to two alleles, one inherited from each parent. On an electropherogram, these alleles show up as spikes, and vary from person to person. We usually share half or more of our markers with close relatives, and often share several with complete strangers. But by the FBI’s statistics, the probability of sharing all 13 loci with someone you’re not related to is lower than 1 in a trillion.

The cheese wrapper had only produced a partial profile; spikes showed up at just six of 13 loci. This didn’t surprise Fahnestock. As DNA ages, it breaks down. Some markers are famous for crumbling into illegibility right away; others can be detected decades later. All things considered, six loci seemed reasonable for a sample from the 1980s.

But the partial profile put Fahnestock in a quandary. At the time, he didn’t know whose DNA he had found, so he wanted to search the federal database for the profile. That database, called the Combined DNA Index System, or CODIS, contains more than 14 million profiles, including all convicts and many arrestees from the state of Missouri for the last 15 years. But it cannot be searched without entering at least ten of the 13 loci, and Fahnestock only had six. Fahnestock asked for permission to make an exception and search for the partial profile — an uncommon but not unheard of request. After some discussion and paperwork, the state database administrator agreed.

When Fahnestock ran the cheese wrapper through the database, it returned 10 candidates — all ten were technically “matches” for the profile found on the cheese wrapper. Though we think of a DNA match as unambiguous, partial profiles have so little genetic material that they can result in several potential matches, and analysts must interpret which match looks best. Fahnestock did just that and then disregarded the nine other matches.

A few days later, he learned the identity of that match: Brian McBenge.

The defense team argued that because the profile was incomplete, it was more likely to be a false positive — a hit by unhappy happenstance.

It’s not clear how often coincidental matches occur. The FBI has argued that it’s rare, but some statisticians disagree, as a 2008 investigation by the Los Angeles Times revealed. The newspaper wrote that a rogue Arizona state employee had run tests on the state’s database without the FBI’s permission and found 122 pairs of profiles that matched at 9 or more loci. Twenty of them matched at 10 loci. One pair matched at 11 and another at 12. This all happened in a database with just 65,493 profiles.

The Arizona results were not anomalous: In Illinois’s database of 220,000 profiles, a search found 903 pairs that matched at nine or more loci. Bureau experts say some matches can be expected in a large database, and that others may be close relatives or accidental duplicates. But they have halted further investigation of their statistics, citing privacy concerns.

Brian McBenge in January at the Fulton Reception and Diagnostic Center in Missouri, where he and Cecil were housed for several months after their conviction.

Database matches are tricky. They can sometimes solve otherwise inscrutable cases, but they can also lead investigators down the wrong path. In one case in Bolton, England, police deduced a 6-loci profile from blood discovered on the window of a burglarized home. The trouble was, the DNA matched a man with advanced Parkinson’s disease who could barely walk. But the match statistic — 1 in 37 million — seemed so definitive that police arrested him anyway. He was finally vindicated after more advanced DNA tests revealed that he shared a partial profile with the culprit.

When a suspect has been discovered because of a database match, courts must decide how much weight to give that evidence, and different statistical methods can arrive at wildly different results. The method used by Fahnestock calculates how often the profile is expected to occur randomly in the population. After his initial tests, he calculated that the probability of a random person having the profile he found on the cheese wrapper was 1 in 741,000 among Caucasians.

Defense expert Dan Krane came up with a different statistic: 1 in 2.5.

The reason for the yawning difference, Krane explained, is that his method takes into account the likelihood of a coincidental match in a specific database. For instance, if you search a database of 1 million for a profile that 1 in 100,000 people share, you would expect around ten hits. If you arrested any of them without other evidence, you’d probably get the wrong person. The method was endorsed by a special advisory group to the FBI and a National Research Council panel, but is sometimes withheld from court in an attempt not to confuse jurors with statistical arguments.

The McBenge team contends that because the DNA on the cheese wrapper was degraded, it was far more likely to produce a false positive than a full profile would have. If investigators had found substantial other evidence pointing to their guilt, the database hit, even one with a 1 in 2.5 chance of being reliable, would have been the cherry on top. But without that substantial evidence, the juries were left to decide how much faith to place in the DNA.

I n contrast to the minimal spikes on the cheese wrapper’s electropherogram, the stocking’s chart was a mountain range. There could be only one explanation: Genes from multiple people were on the nylon. One of those people appeared to be Knoernschild, which was a good sign for the prosecution’s theory. The stocking, though found in the back alley, had probably come from her house.

Identifying who else was on the stocking was more challenging — particularly because the DNA there was degraded, so only some markers could be detected. Fahnestock approached the problem by puzzling out the spikes that didn’t belong to Knoernschild and creating a theoretical profile of a second contributor. Because that profile had at least one matching gene at 10 loci, Fahnestock didn’t need special permission to run it through the federal database. The result matched Cecil McBenge.

In forensic science, interpretation of mixtures is a known danger zone. When a sample has more than one person’s DNA, as the stocking did, the analyst must interpret the data, making decisions about how many individuals may be in the mix, and which spikes belong to which person. When people in the sample happen to have similar genes, it gets more complicated. Cecil and Knoernschild shared alleles at seven of their 13 loci.

A 2013 survey by the National Institute of Standards and Technology asked analysts from 108 labs to look at a three-person mixture and determine if a suspect’s DNA was present. Seventy percent of the analysts said the suspect might be in the mix; 24 percent said the data was inconclusive. Just six percent arrived at the truth: The suspect was not in the sample.

Not only do analysts vary in their interpretation of evidence, they also disagree over how certain to feel about the results. In another NIST survey, labs interpreting a two-person mixture came back with match probabilities that varied by 10 orders of magnitude. “Imagine if you take a pregnancy test and you send it to two different labs,” said Greg Hampikian, who authored the study on bias in the Atlanta rapist case, “and one said the odds are a billion to one that you’re pregnant, and the other said it’s 50-50.”

B ecause different analysts can reach different conclusions about the same DNA evidence, savvy investigators shop evidence around to get the results they want. In one instance, the California Innocence Project had a large private lab test evidence attorneys believed could exonerate a convicted murderer in Los Angeles. That lab reported the DNA was inconclusive. So lawyers took the same data to another analyst. By her assessment, the evidence plainly cleared the convict.

California Innocence Project attorney Mike Semanchik said he had no qualms about asking multiple analysts to look at the same data — doing otherwise would be a disservice to his client. Nonetheless, the implications are troubling, he said. “How can two labs get entirely different answers from the same DNA tests?”

Even the most automated part of the process — the analysis performed by machines — varies from lab to lab. Among other tasks, forensic profiling equipment amplifies DNA to an observable level. Most labs amplify DNA to the maximum the manufacturer recommends. But when that isn’t enough to produce legible spikes, some labs will hike up amplification or otherwise futz with the settings to get results.

Methods that push equipment beyond its normal bounds are commonly called low copy number DNA testing, or LCN testing. This is the most controversial practice in forensic biology today because of the errors it can produce. Pushed too far, the results will begin to show spikes where there are none, while other spikes disappear.

In New York City, the Office of Chief Medical Examiner has embraced LCN; it is the only public forensics lab in the country to have done so. Lab officials say they only use the technology in ways that are scientifically valid and reliable, and the New York State Commission on Forensic Science sanctioned the method in a series of contentious votes.

Last fall, commission member Barry Scheck voiced his concern about the method at a hearing of the DNA subcommittee. Before Scheck made his name disputing DNA as O.J. Simpson’s lawyer, he founded the Innocence Project, which has used DNA to exonerate hundreds of wrongfully convicted people. He said he opposes the use of cutting-edge DNA forensics in court because he doesn't think they have been sufficiently proven. Scheck had been demanding the Office of Chief Medical Examiner make public its internal validation studies on LCN, which it has refused to do. At one point, after an otherwise subdued hearing, he yelled to some of the subcommittee members: “YOU ARE ALL FUCKING LYING!”

San Francisco resident Gale Joseph Young spent years incarcerated while lawyers fought over LCN testing. Young had been questioned and strip-searched in a San Francisco police station on suspicion of selling drugs in 2008. Officers found nothing on him, but after releasing him, they saw a plastic baggie with 14 grams of crack cocaine on the floor. There was so little DNA on the bag that the crime lab used LCN to get results, and concluded that he might be in the mix.

The Fulton Reception and Diagnostic Center.

Despite defense objections, these results were allowed in court. One jury hung; a second convicted. The case was appealed to U.S. District Judge Maxine Chesney, who was tasked with deciding the LCN question. Chesney, befuddled by the technical arguments, ruled the LCN evidence was admissible because neither side was “being ridiculous in their general approach.” The case was appealed to the Ninth Circuit Court, which reversed her decision and admonished her for failing to adequately evaluate the science (“Not being ridiculous is not synonymous with being reliable.”).

Unfortunately for the clarity around LCN, neither Chesney nor the higher court ruled on the validity of the science itself, because by then, Young had already served a 5-year, ten-month sentence, and prosecutors dropped the charges.

Fanhestock says his lab avoids such controversy by staying away from LCN testing altogether. If a sample has less DNA than his equipment has been validated for, he won’t blow it up with more amplifications.

But defense attorney and DNA expert Bicka Barlow countered in depositions and in court that his lab did toy with LCN in the McBenge cases: She noted that in some tests, he had doubled the quantity of DNA and increased testing time in order to heighten the low spikes. These techniques, she said, could result in errors. Fahnestock maintained that his methods were valid and uncontroversial and his results were reproducible.

To see if they were, seven analysts agreed to participate in an informal experiment for this story. This experiment, though hardly randomized or double-blind, would hopefully yield insights about the precision of DNA forensics. The participants studied the original electropherograms and accompanying data from the cheese wrapper and stocking in Knoernschild’s case, and deduced the DNA profiles on them. When the results came back, Dr. Lawrence Kobilinsky, chair of forensic science at John Jay College of Criminal Justice in New York, scrutinized them for variance. It wasn’t hard to find. Two of the seven respondents wouldn’t do the analysis as directed, because they were instructed to go about it exactly as Fahnestock had — without a stochastic threshold.

A stochastic threshold is a way to anticipate problems: The threshold is established by testing equipment with shrinking amounts of DNA until the results are no longer accurate. Anything below the threshold is unreliable. As of 2012, about 86 percent of labs used a stochastic threshold. But in 2011, Fahnestock was still using equipment for which a stochastic threshold had never been established. (In 2014, Fahnestock retested a different patch of the stocking on newer equipment that did have a stochastic threshold, but he couldn’t do the same for the cheese wrapper, because all the DNA had been used in the initial testing.)

Even those analysts who participated arrived at different results, particularly in their certainty statistics. For the cheese wrapper, Fahnestock had ascribed a probability of 1 in 741,000 of a random match. The seven analysts’ rates spread from 1 in 3,670 to 1 in 593,000 (a full profile would have resulted in a statistic closer to 1 in 1,000,000,000,000,000).

The stocking provided even less consensus. Fahnestock had assumed a mixture of two people — Cecil and Knoernschild. One analyst concluded it was a mix of at least three. Another came up with a result that excluded Cecil entirely.

Again, the certainty rate was wholly uncertain. Fahnestock’s final analysis ascribed a probability of 1 in 6.17 million of a random match; the analysts’ stats ranged from 1 in 4,470 to 1 in 106 billion. “I’m kind of astonished that there would be that much variation,” Kobilinsky said. “They all have the same job, and they’re coming up with different results.”

T he trial of Brian McBenge began in August 2014; his brother’s trial began six weeks later.

They were each charged with first-degree murder, so the prosecution had to prove they’d killed with intent. Prosecutors maintained that Brian had burglarized Knoernschild’s house in 1980, and returned four years later with his brother in search of more loot. This time, prosecutors said, Knoernschild was at home, so they killed her.

Brian was neither questioned nor charged at the time of the 1980 burglary. But Debbie Wines testified that she now remembered breaking up with Brian before that burglary, so there would be no legitimate reason for his fingerprint to be on the refrigerator. The prosecution showed the jury side-by-side photographs of the 1980 and 1984 ransackings. The same cupboards were opened, food dumped on the floor. Brian had motive and opportunity in both instances, the prosecutors told jurors, because he knew about the cash-filled Calumet can.

The defense teams attacked this evidence; the fingerprint was only partial, and Debbie’s memory of when she and Brian broke up was hazy at best, they said. The crimes had happened so long ago that neither Cecil nor Brian could provide an alibi. And there was no evidence that anything had been stolen from the house in either burglary: In 1980, Knoernschild hadn’t reported any money missing; in 1984, the can was found apparently untouched in the back of Knoernschild’s closet. Money was also found in her purse.

But neither a fingerprint nor a can of cash were likely to return a verdict. The outcome depended on the DNA found on the cheese wrapper and stocking. For a conviction, the jury would have to trust the DNA. For an acquittal, the jury would have to believe that DNA forensics can arrive at the wrong answer.

The prosecution blew up the electropherograms to poster-size and explained each spike to the jury. The defense attacked the relevance of those spikes: The DNA was found on items not intimate to the crime. Fifteen fingerprints were lifted from her house after she was killed — none were traced to Cecil or Brian. If they had committed the crime, Cecil’s attorney Cyndy Short said, how could they leave so little of themselves behind?

The prosecution countered. How likely was it that two brothers’ DNA could accidentally arrive on two pieces of evidence from the same crime scene?

The defense walked the jury through statistical and technical arguments to describe how that may have happened. The DNA was degraded, mixed, contaminated, meager and partial — a complete catalog of the danger zones in DNA analysis. The gap between good science and infallible science is wide, Short argued, and left ample room for reasonable doubt.

The totality of the state’s evidence against Cecil and Brian, Short concluded, was one partial fingerprint from the 1980 burglary, the photos comparing the crime scenes from 1980 and 1984, and DNA on a cheese wrapper and a knee-high stocking.

But it was enough.

The jury took four hours to convict Brian McBenge. He received a life sentence. A month and a half later, another jury delivered the same verdict to Cecil.

The judge in Cecil’s trial, Daniel Pelikan, could not comment on the specifics of the case, but said his juries have typically found DNA evidence extremely persuasive. “In all my cases, I warn the juries that the trials take longer than an hour to complete, and not every case has fingerprints, saliva, semen, and DNA,” he said.

But, Pelikan added, “We have a well-educated, sophisticated jury pool in my county. I’ve always been impressed with their ability to understand the science.”

The McBenges are now appealing their convictions on more than a dozen points. Cyndy Short is optimistic about the chain of custody question: That the cheese wrapper wasn’t checked into evidence for nine days after Bob Brockmeyer collected it, for instance, and the stocking arrived to the crime lab for testing in an unsealed envelope. The defense team is less hopeful about their appeal on the strength of the evidence; DNA evidence does not tend to inspire doubt.