Heinrich’s Pyramid in 2026: What Modern EHS Teams Actually Use It For
The safety pyramid arranges workplace incidents by severity: a single serious injury at the apex, a larger band of minor injuries below it, and a wide base of near misses and unsafe conditions supporting both. Nearly every safety induction in manufacturing has shown a version of it. Its longevity is remarkable given how little scrutiny it survived. The pyramid is now roughly ninety five years old, has been revised once with materially different numbers, and has been the subject of a research finding that undermines the use most sites make of it. Yet it still appears on wall boards, largely because the alternative to a memorable image is usually no image at all. The model deserves neither the reverence nor the dismissal it tends to receive. It is worth knowing precisely what it claimed, what held up, and which parts of a modern safety programme it can honestly support. A Model Assembled From Insurance Claims Herbert William Heinrich worked in the engineering and inspection division of an insurance company, which gave him access to a large volume of accident claim files. In 1931 he published an analysis of them, and two findings from it entered the profession permanently. The first was a ratio. For every major injury, Heinrich counted 29 minor injuries and 300 accidents that caused no injury whatsoever. Drawn as a triangle, this became the image everybody knows. The second finding travelled further and did more harm. Heinrich attributed 88 per cent of accidents to the unsafe acts of workers, 10 per cent to unsafe mechanical or physical conditions, and 2 per cent to the unpreventable. That single percentage licensed decades of safety programmes aimed squarely at operator behaviour, poster campaigns about carelessness, and investigations that stopped at the person nearest the machine. Both figures now rest on nothing that can be examined. The claim files Heinrich reviewed were never published and no longer survive. A profession reproduced his ratio for ninety years without anybody being able to check it. Bird Repeated the Exercise and Got Other Numbers Frank Bird ran a much wider study in 1969, drawing on well over a million incident reports from companies across a broad span of industries. He found the same general shape and quite different proportions. Heinrich, 1931 Bird, 1969 Serious injury 1 1 Minor injuries 29 10 Property damage Not recorded 30 Near misses 300 600 Read those columns against each other, because they contain the most useful lesson the pyramid has to offer. Two competent analyses of large incident datasets differed by a factor of three at the minor injury band and doubled the base. Bird also introduced a category, property damage, that Heinrich had not counted at all. Ratios describe the population they were drawn from. They shift with industry, with era, with what a company chose to record, and with how honestly its workforce reported. A pharmaceutical site that expects to find exactly 300 near misses under each serious injury has mistaken an average from another century for a property of the physical world. The Question the Pyramid Never Answered Look closely at what the model actually asserts. It sorts events by severity and counts them. That is all. Somewhere in the retelling, this became a causal claim: that the small events at the base produce the large ones at the apex, and that grinding down the base must therefore shrink the top. Neither Heinrich nor Bird demonstrated this, and a correlation between the layers would arise anyway from a simple fact, which is that a plant exposing people to more hazard generates more of every category of incident. Suppress the near misses and a site suppresses its near misses. Whether the fatality risk moves at all depends on something the pyramid is silent about, namely whether the events at the top and the bottom arise from the same conditions. Injury Rates Fell. Fatalities Held Steady. That silence turned out to matter enormously. Through the 2000s a pattern surfaced across heavy industry, and once named it was recognised almost universally. Recordable injury rates had declined year after year under sustained behavioural safety effort. Fatality rates had barely shifted. Organisations celebrating a decade of improving lost time injury figures were killing people at approximately the rate they always had. The explanation is now widely accepted. Minor injuries overwhelmingly come from routine, low energy exposures: a slip on a wet walkway, a strain lifting a drum, a laceration from a sharp edge on a guard. Fatalities come from high energy exposures and from the failure of controls meant to be absolute. Working at height. Stored electrical or hydraulic energy. Vehicle and pedestrian interaction. Confined spaces. Uncontrolled release of chemical or thermal energy. A wet floor and a defeated press interlock are both incidents. Beyond that they have almost nothing in common, and no amount of attention to the first will discover the second. A site can watch its injury frequency rate fall for four consecutive years while walking steadily toward a fatality, and its safety dashboard will show green throughout. What Actually Survives Three uses hold up, and they are narrower than the model’s reputation suggests. The base contains a site’s only free information. Near misses and unsafe conditions are the sole category of incident that arrives without a cost attached, and they arrive in quantity. That is a strong argument for capturing every one of them, and it says nothing about ratios. Severity as recorded is a bad organising principle, and severity as possible is a good one. The spanner dropped from a gantry is the same event whether it lands on concrete or on somebody’s shoulder, and chance decides which. Sorting the base by what could plausibly have happened, rather than by what did, pulls the handful of events that share causes with a fatality out of the ninety that describe housekeeping. That sorting only works if every supervisor on every shift applies the definitions identically, which makes data standardisation the precondition for the entire exercise