Biological interpretations of a competing hazards model intended to express the age pattern of mortality throughout the life span are evaluated using cause of death data. The model examined is the Siler model, which consists of three competing hazards, immature, residual, and senescent. The data employed are the worldwide sample of life tables and decremented life tables, that is, life tables with a cause of death eliminated, assembled by Preston et al. (Causes of Death Life Tables for National Populations. New York: Seminar Press, 1972). The results indicate that causes of death are accounted for by the model in a manner consistent with the biological interpretations of the hazard functions. For example, the cardiovascular diseases, neoplasms, and other degenerative diseases are accounted for by the senescent component of morality. This is a valuable characteristic of the model that may prove useful for assessing errors in cause of death data and identifying variation in the impact of a cause of death on the age patterns of mortality. For example, only 30% of deaths classified as due to other and unknown causes are attributed by the model to the senescent component of mortality and are consequently likely to be due to degenerative diseases. This finding contradicts the conventional assumption that the majority of other and unknown causes of death are due to the degenerative diseases. Additional results show that certain causes of death, such as influenza, pneumonia, and bronchitis, can occur in different components of mortality, suggesting that some categories of death may express several different etiologies. Partial correlation analysis indicates that much of this variation in etiology occurs among (and not within) populations. It is concluded that the Siler model is a useful framework for studying historical and/or crossnational trends in mortality.A number of models have been proposed that account for changes in age-specific mortality rates over the entire human life s an.ards functions (Theile, 1871;Siler, 1979; Heligman and Pollard, 1980;Mode and Busby, 1982;Mode and Jacobson, 1984) and the relational methods (Brass, 1981;Ewbank et al., 1983). The hazard models are nonlinear mathematical functions intended to represent the age related changes in the risk of dying, also known as the hazard function, the instantaneous death rate, or the force of mortality. The relational methods, on the other hand, obtain the basic shape of the human mortality curve from a standard tabular life table and then use linear re ession to fit the logit transform of the stan f-ard survivorship curve to the logit These models are often divided into the R aztransform of the observed survivorship curve. Both classes of models are useful for graduatin the a e-s ecific mortality rates of a life ta % le an C i W or or summarizing a life table by a smaller number of parameters.One advanta e of the relational models is that they ten % to contain fewer parameters than the hazard models and are consequently easier to fit to data. Th...