Dynamic mathematical modelling and stochastic simulation of disease–host systems for the purpose of epidemiological analysis offer great opportunities for testing hypotheses, especially when field experiments are impractical or when there is a need to evaluate multiple experimental scenarios. This, combined with the ever increasing computer power available to researchers, has contributed to the development of many mathematical models for epidemic simulations, such as the individual-based model (IBM). Nevertheless, few of these models undergo extensive validation and proper assessment of intrinsic variability. The Ontario rabies model (ORM) will be used here to exemplify some advantages of appropriate model behaviour validation and to illustrate the use of a simple geometric procedure for testing directional bias in distributed stochastic dynamic model of spread of diseases. Results were obtained through the comparison of 10 000 epizootics resulting from 100 epidemic simulations started using 100 distinct base populations. The analysis results demonstrated a significant directional bias in epidemic dispersion, which prompted further verification of the model code and the identification of a coding error, which was then corrected. Subsequent testing of the corrected code showed that the directional bias could no longer be detected. These results illustrate the importance of proper validation and the importance of sufficient knowledge of the model behaviour to ensure the results will not confound the objectives of the end-users.