We Investigated How Accurate Forensic Methods Are. Here's What We Found.
Veritasium
An Ancient Case With Flies and Sickles 0:00
A 1247 book called The Washing Away of Wrongs by Song Ci describes an early empirical approach to forensics. A man is found dead with wounds from a curved blade, so a village official orders every farmer to lay down his sickle in the square. Flies drawn to trace blood settle on a single sickle, and its owner confesses. The story is framed as the first written example of using observation to solve a crime, even though something as simple as swapping tools could have defeated it.
Modern Forensics Under Scrutiny 0:30
Fingerprints are said to support over 70 percent of murder investigations and DNA evidence more than 90 percent, yet real cases show the risks, like a hair sample that sent men to prison for 28 years before it turned out to be dog hair. In 2009 the National Academy of Sciences published a 350 page report concluding that, apart from nuclear DNA analysis, no forensic method has been rigorously shown to reliably link evidence to a specific person, and that some techniques do not even meet basic scientific standards. People were then asked to rank five well known forensic techniques from least to most accurate, and most were unsure where to place them.
Hair Analysis Falls Apart 2:00
Before DNA testing existed, investigators compared the microscopic surface and pigment patterns of hairs to try to match suspects to crime scenes. Between the 1970s and 1999 the FBI used this method in 268 cases, but when DNA testing was later applied to the same evidence, 96 percent of those hair matches were found to be false. Examiners sometimes could not even distinguish human hair from dog hair. Thirty three people from these cases had already been sentenced to death, and nine had already been executed before the errors were uncovered, which is why hair analysis ranks as the least accurate method on the list.
Bite Marks Never Passed the Test 3:30
Bite mark analysis assumed that if dental records could identify a body, the same logic could identify an attacker from marks left on a victim, and it was used in courts from the 1950s onward. Forensic dentist Mary Bush later pointed out that no real scientific testing was done before the method was accepted in court. A University at Buffalo study created 89 bite marks in cadavers and compared them to marks made in wax by the same model teeth, and none of the skin marks matched. When compared against 411 different sets of model teeth, the original teeth were not even the closest match. Despite twelve studies concluding that skin is too soft and distorts too much to reliably hold a bite mark, the method still has not been banned and was used in court as recently as 2025.
Blood Spatter Improves Over Time 5:00
Bloodstain pattern analysis interprets how blood pools and splashes to reconstruct a crime. It began in 1971 when chemist Herbert Leon MacDonell published a book explaining how the width and length of a stain could be used with trigonometry to trace back to where the blood came from. Courts, including the Supreme Court of Iowa, accepted the method without demanding proof of its reliability. The trouble is that MacDonell's original calculations ignored gravity and drag, which shift the calculated origin point, for example wrongly suggesting a victim was standing when they were actually sitting. The first study measuring its baseline reliability only came in 2014, and a 2021 study found analysts disagreed on how a stain was made about 8 percent of the time, partly because blood itself varies between people, with men averaging 15 percent higher blood cell concentration than women. New software using fluid dynamics is improving accuracy today, placing this method in the middle of the ranking.
Fingerprints and the Mayfield Case 8:02
After the 2004 Madrid train bombings that killed 193 people, a fingerprint on a detonator bag was matched to Oregon lawyer Brandon Mayfield, even though he had no passport and had not left the United States in over a decade. The modern fingerprint system traces back to 1890s Kolkata, where officers Edward Henry, Azizul Haque and Hem Chandra Bose built a classification system based on whorl patterns and other features, sorting prints into 1024 wooden pigeonholes, a system later used for a century at Scotland Yard. Matching a print still requires identifying features called minutiae, but research shows different examiners can spot anywhere from three to ten minutiae on the same print, and the same examiner can reach a different conclusion on the same pair of prints 10 percent of the time. Context bias makes this worse, since as many as 42 percent of fingerprint requests tell examiners whether a suspect has a criminal record, and in Mayfield's case even the defense's own expert agreed with the false match, an example of conformity bias.
DNA Has Limits Too 16:33
DNA analysis has become so sensitive that a 61 year old cold case was recently solved using just 0.4 nanograms of genetic material, but that same sensitivity creates new risks. In 2012 paramedics in California accidentally transferred a homeless man's DNA from their gloves onto a murder victim, leading to his arrest and five months in jail before charges were dropped. DNA left on shared surfaces like door handles often mixes profiles from several people, and a 2013 study by the National Institute of Standards and Technology found that 69 percent of labs misinterpreted a mixture from four contributors, with only 21 percent correctly calling it inconclusive. There are still no fixed standards for how much or how mixed a sample can be before a lab should refuse to analyze it.
Where That Leaves Forensic Science 22:00
The video closes by saying its aim was not to discredit forensics, since a world with these methods is still better than one without them. It argues that keeping forensics scientific requires continuing the work of researchers who have spent years reassessing and correcting these techniques. One expert defends calling it science, saying it means doing the best possible with the evidence at hand until something more reliable comes along.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
