This paper is devoted to the issue of estimation of cause-specific risk of death. The traditional methodology is compared with one based on the empirical Bayes approach to statistical inference. The presentation of both methods and the evaluation of their theoretical properties is illustrated by an empirical example of how they serve to assess the respective situations in the Czech Republic, the Netherlands and Poland from 1994-1996. The results confirm earlier observations (e.g. Clayton and Kaldor, 1987) that the empirical Bayes estimates of the relative risk of death are less extreme and less dispersed than the traditional epidemiological measures (i.e. Standardized Mortality Ratios). Improvement of accuracy of estimation in the Bayesian approach results from including additional prior information on the phenomena under study. It is argued that due to the aforementioned features, the Bayesian methodology is suitable for the study of small-sample populations, insofar as it provides less extreme and more precise estimates, than the traditional methods. This is an important issue in studies of regional mortality profiles, where insufficient sample size is often a serious problem. At the same time, the philosophy of Bayesian statistics, and particularly the subjective definition of probability, is a natural premise of analysis in the case of mortality research, where the samples of events under study are not repeatable. Therefore, although the Bayesian approach is more complex (both philosophically and procedurally), it constitutes a valuable alternative method of cause-of-death mortality risk evaluation, especially at the regional level, where small samples are considered.