The reliability of different statistical road accident severity models is assessed using a number of criteria, such as goodness of fit, robustness of risk factor coefficients, and intuitive acceptability and consistency of output. The results of this analysis suggest that model reliability is not sensitive to the number of injury classes specified in the model or to the level of model aggregation. All models consistently identified the same risk factors as explaining most of the variation in injury experience in the data. These factors are (a) dynamics of the accident, (b) seating position of occupant, (c) use of seat belts, and (d) age of occupant involved. There is no indication that a significant transfer of error takes place from one severity level to another in a sequential model structure. The results suggest that reliability of statistical road accident severity models depends primarily on the accuracy of information provided in the accident data.