When the fire is out, the hardest question begins: where did it start, and why? A fire investigator has to reconstruct an event that destroyed much of its own evidence — and do it in a way that stands up in court. Two words frame the whole job: origin and cause.

Origin before cause

You cannot reliably determine what caused a fire until you know where it began. Fire generally burns upward and outward, so investigators work backward from the least-damaged areas toward the most heavily burned, reading the building like a record of the fire's travel.

Reading burn patterns

Fire leaves patterns that, read carefully, point toward the origin:

  • V-patterns on walls — the point of a "V" of soot and damage often sits above the origin.
  • Char depth — how deeply wood has burned can indicate where the fire burned longest.
  • Calcination of plaster, and the condition of materials, glass and furnishings.
  • Low, localised damage that doesn't fit the natural upward spread.

No single pattern is decisive; the investigator weighs them together.

Finding — and ruling out — a cause

Once the origin area is identified, the investigator looks for what ignited the first fuel: electrical faults, discarded smoking materials, cooking, heaters, or deliberate ignition. Crucially, good practice means systematically eliminating accidental causes rather than leaping to arson.

Where an accelerant is suspected, investigators look for irregular "pour" patterns and use vapour detectors, trained accelerant-detection dogs, and laboratory gas chromatography–mass spectrometry (GC-MS) on debris samples to confirm the presence of ignitable liquids.

Discipline against bias

Modern fire investigation is built on the scientific method and guidance such as NFPA 921: form a hypothesis about origin and cause, then test it against the evidence and actively try to disprove it. This matters because several past "arson" convictions were later overturned when the burn patterns they relied on were shown to be misread. Honest investigation means being willing to conclude "undetermined" when the evidence doesn't support more.