This study presents a description of the concept of behavioural culpability, a step-by-step manual for using it, and an empirical test of a suspected mis-classification of culpability. Behavioural culpability is defined as whether the driver’s actions contributed to a crash and that non-culpable crashes are not caused by any specific behaviour and can only be predicted from exposure. Drivers with non-culpable crashes are therefore a random sample of the population. However, if the criteria for culpability and/or the individual judgements are not reflective of the principle of behavioural culpability, no fault drivers will not be a random sample of the driving population. To test the predictions from the definition of randomness in a sample assumed to have sub-optimal coding, the categorization of crash involvement undertaken by a British bus company was tested for associations between at fault and no fault crashes, age and experience. As predicted from the low percentage of at fault accidents in the sample, correlations between the variables indicated that a fair percentage of at fault crashes had been coded as no fault of the bus driver, suggesting a too lenient criterion. These results show that within fleet-based companies, culpability for a crash is probably allocated for legal reasons, which means that the predictability of accident involvement taking into account individual differences is not fully utilized. The aim of behavioural culpability coding is to increase effect sizes in individual differences in safety research and to improve our capability of predicting accident involvement.