The epidemiological technique known as the ‘found experiment’ was used to investigate whether or not an industrial sample showed non‐random differentiation with respect to individual accident rates. For the particular sample ‘found’, it was possible to control for such conditions as differences in the jobs being done and the environment in which they were undertaken, the period of exposure to risk, and the age and experience of the employees. Non‐random differentiation was found within the sample over the period of 8 years and 9 months for which accident records were available, in that the distribution of accidents over the sample did not differ significantly from the negative binomial distribution. This differentiation was also found to be stable over time, since the correlation between individuals' accident rates in the first and second halves of the period was + 0.67. Implications for future accident research of these findings and of the use of the ‘found experiment’ technique