On February 17, 2004, Texas executed Cameron Todd Willingham for setting a fire that killed his three young daughters. Thirteen years earlier, fire investigators had walked through the charred remains of his Corsicana home and seen what they believed were unmistakable signs of arson: strange patterns burned into the floor, cracked glass in the windows, deep charring that could only mean accelerants had been used. They identified over 20 indicators that, to their trained eyes, proved Willingham had murdered his children. Every one of those indicators has since been debunked.[s] The state of Texas almost certainly executed an innocent man, and the flawed arson investigation science that convicted him had been sending innocent people to prison for decades.
The Folklore of Fire
For most of the 20th century, arson investigation science was less science than folklore. Investigators learned their craft through apprenticeship, watching experienced colleagues read fire scenes the way fortune tellers read tea leaves.[s] They looked for specific indicators that supposedly proved a fire had been set deliberately: “crazed glass” with its web of fine cracks, “pour patterns” on floors, deeply charred wood, melted metal, and collapsed furniture springs.
None of these methods had ever been scientifically tested. A 1977 government report noted that common arson indicators had “received little or no scientific testing” and that there was no published scientific literature to substantiate their validity.[s] But the findings went largely ignored. Fire investigators continued teaching these methods to the next generation, perpetuating what one expert would later call “old wives’ tales.”[s]
The Myths Unravel
The revolution in arson investigation science began in 1991, when fire investigators in Jacksonville, Florida, made a startling discovery. They were examining a house fire that bore all the hallmarks of arson: rapid spread, intense heat, pour patterns on the floor. But when they set up a controlled burn in an identical vacant house next door, they watched an accidental fire produce exactly the same indicators.[s]
The culprit was a phenomenon called “flashover,” the moment when accumulated heat causes everything flammable in a room to ignite simultaneously. In flashover, a fire that started from a dropped cigarette becomes indistinguishable from one started with gasoline. The revelation forced fire scientists to confront an uncomfortable truth: much of what they had been teaching as proof of arson could happen in any sufficiently hot fire.[s]
Crazed glass, once considered ironclad evidence of accelerant use, turned out to be caused by cold water from fire hoses hitting hot windows.[s] Pour patterns were created by melting furniture and falling debris. The supposed indicators of arson were actually just indicators of fire.
Lives Destroyed by Bad Science
Ernest Ray Willis spent 17 years on Texas death row for a 1986 fire that killed two women. Like Willingham, he was convicted based on testimony about burn patterns, heat intensity, and other supposed arson indicators. In 2004, months after Willingham was executed, prosecutors hired a new fire expert to review Willis’s case. His conclusion was devastating: “There is not a single item of physical evidence in this case which supports a finding of arson.”[s] The fire was likely caused by a faulty electrical outlet. Willis walked free, but only after losing nearly two decades of his life.
Han Tak Lee was convicted in 1990 of setting a fire that killed his mentally ill daughter at a religious camp in Pennsylvania. Fire investigators testified about burn patterns and the intense heat of the fire. A federal judge would later describe the case as one where “what was once regarded as truth is myth, and what was once accepted as science is superstition.”[s] Lee served 24 years before his conviction was overturned in 2014.[s]
John Henry Knapp was sentenced to death in Arizona for a 1973 fire that killed his two young daughters. Investigators found burn patterns they attributed to accelerants and multiple points of origin. Knapp came within days of execution before lawyer Larry Hammond used new evidence about flashover to demonstrate that accidental fires could produce identical patterns. Knapp was freed in 1987 after 17 years in prison.[s]
The Scale of the Problem
These cases are not isolated failures of arson investigation science. Misapplied forensic evidence has contributed to more than half of wrongful convictions handled by the Innocence Project.[s] In an NIJ analysis of 45 fire-debris-investigation case examinations drawn from National Registry of Exonerations cases, 78% contained at least one case error.[s]
Gerald Hurst, the fire scientist who reviewed both the Willingham and Willis cases, estimates that at least one-third, and perhaps half, of all arson convictions have been based on junk science. In Texas alone, that could mean 250 to 400 innocent people sitting in prison.[s] What we see in terms of exonerations, one expert notes, “is just the tip of the iceberg.”[s]
The Long Road to Reform
Change has come slowly. In 1992, the National Fire Protection Association published NFPA 921, the first comprehensive guide requiring fire investigators to use scientific methods rather than folklore.[s] The 2009 National Academy of Sciences report on forensic science found that many commonly used techniques, including those in arson investigation science, had never undergone the testing necessary to establish their reliability.[s]
Yet progress remains uneven. The Texas Forensic Science Commission found in 2011 that unreliable fire science methods were still in use, issuing 17 recommendations for reform including mandatory training and new certification criteria for expert witnesses.[s] Some investigators continue to rely on methods that have been discredited for over three decades.
Cameron Todd Willingham cannot be brought back. But the transformation of arson investigation science from folklore to genuine forensics offers hope that fewer innocent people will follow him to prison or the execution chamber. The question now is whether the criminal justice system can learn from its mistakes quickly enough to free those who are still paying the price for old wives’ tales dressed up as expertise.
The arson investigation science used to convict Cameron Todd Willingham in 1992 relied on over 20 indicators that investigators Manuel Vasquez and Douglas Fogg claimed proved he had intentionally set the fire that killed his three daughters. They pointed to burn patterns on the floor that they said showed accelerant had been poured, “crazed glass” with fine cracking they attributed to extreme heat, and charring patterns they believed indicated the fire had burned unusually hot and fast.[s] Days before Willingham’s execution in 2004, fire scientist Gerald Hurst submitted a report concluding that the indicators were invalid; the state let the execution proceed anyway. After Willingham was put to death, the Innocence Project assembled five of the nation’s leading independent arson experts, who issued a 48-page report finding that none of the scientific analysis used to convict him was valid.[s]
The Methodology Problem
The fundamental problem with traditional arson investigation science was its reliance on pattern recognition without controlled experimentation. Investigators observed certain characteristics at fire scenes where arson was suspected, and those characteristics became codified as “arson indicators.” But the scientific method requires testing hypotheses against alternatives. No one had conducted systematic studies comparing intentionally set fires to accidental ones.
A 1977 government report conducted by the Law Enforcement Assistance Administration documented this gap, finding that common arson indicators had “received little or no scientific testing” and noting the absence of any published scientific literature validating these methods.[s] The report’s findings were largely ignored, and the folklore continued to be taught as fact.
Debunked Indicators
Crazed Glass: Windows with fine weblike cracking were long considered proof of extremely rapid heating, which investigators attributed to accelerant use. Controlled experiments demonstrated that crazed glass actually results from thermal shock when cold water from fire hoses contacts hot glass.[s] The indicator says nothing about how a fire started.
Pour Patterns: Irregularly shaped burn patterns on floors were interpreted as showing where an arsonist had poured flammable liquid. Research revealed that these patterns have multiple innocent causes, including melting furniture, falling debris, and the phenomenon of flashover, in which accumulated heat causes all combustible materials in a room to ignite simultaneously.[s]
Spalled Concrete: Pitting and discoloration in concrete were attributed to accelerant exposure. Testing showed that spalling indicates only that concrete became very hot, which can occur in any fire regardless of cause.[s]
Alligatoring: Shiny, textured char patterns on wood were said to indicate accelerant use. These patterns actually result from the natural combustion characteristics of wood and provide no information about fire origin.[s]
The Jacksonville Experiment
The turning point came in 1991 when investigators in Jacksonville, Florida, examined what appeared to be a clear arson case. The fire had spread rapidly, and pour patterns were visible on the floor. But fire scientist John Lentini and investigator John DeHaan obtained permission to conduct a controlled burn in an identical vacant house next door. Using only furniture and household materials, they ignited a fire from a dropped cigarette on a couch.
The test fire reached flashover in roughly four minutes. When examined afterward, it showed pour patterns, deep floor burning, and multiple apparent points of origin, all indicators that would traditionally be classified as proof of arson.[s] The revelation demonstrated that accidental fires reaching flashover are visually indistinguishable from intentionally set fires. Harvard University and the National Bureau of Standards had filmed the flashover phenomenon in the mid-1980s, but the Jacksonville experiment showed its direct implications for arson investigation science.[s]
Case Evidence Review
Cameron Todd Willingham (Texas, executed 2004): Fire expert Craig Beyler’s review for the Texas Forensic Science Commission found that investigators Fogg and Vasquez failed to consider alternative explanations including electrical causes or accidental ignition by the children, and that neither investigation met current scientific standards.[s]
Ernest Ray Willis (Texas, exonerated 2004): Willis spent 17 years on death row based on testimony about “pour patterns” and heat intensity. When prosecutor Ori White commissioned a new fire analysis in 2004, expert Gerald Hurst concluded: “There is not a single item of physical evidence in this case which supports a finding of arson.”[s] The fire was most likely caused by an electrical problem, a broken ceiling fan or faulty outlet.[s]
Han Tak Lee (Pennsylvania, exonerated 2014): Lee was convicted in 1990 based on fire marshal testimony about burn patterns and fire intensity. Federal Magistrate Judge Martin Carlson’s review described the expert testimony as “premised on unreliable science and was therefore itself unreliable.”[s] Lee served 24 years before release.[s]
Current Standards and Ongoing Problems
The National Fire Protection Association’s NFPA 921, first published in 1992, established the scientific method as the required framework for fire investigation. The guide specifies seven sequential steps: identifying the problem, defining the problem, collecting data, analyzing data, developing a hypothesis, testing the hypothesis, and selecting a conclusive hypothesis.[s] The International Association of Arson Investigators formally endorsed NFPA 921 in 2000.[s]
The 2009 National Academy of Sciences report concluded that except for nuclear DNA analysis, many commonly used forensic techniques had not undergone necessary validity and reliability testing.[s] National Institute of Justice analysis of 45 fire-debris-investigation case examinations from National Registry of Exonerations cases found that 78% contained at least one case error.[s]
Despite these reforms, the Texas Forensic Science Commission found in 2011 that unreliable methods remained in use and issued 17 recommendations for improvement, including mandatory adoption of national standards, advanced training requirements, and enhanced certification criteria for expert witnesses.[s]
Expert John Lentini estimates that several hundred innocent people may still be incarcerated for arson-related offenses based on debunked arson investigation science.[s] Gerald Hurst contends that one-third to one-half of all arson convictions may have been based on invalid evidence.[s] The National Registry of Exonerations has documented at least 94 people exonerated after being convicted of arson or related charges, based on evidence that the fire was not arson or that the exoneree did not commit it.[s] For those still in prison, the challenge remains proving innocence in cases where the physical evidence has literally gone up in smoke.
