Knowledge of the differences between deaths for which autopsies are performed and all deaths is pertinent to epidemiologic investigation in such areas as disease trends, chronic disease incidence, and occupationally related disease. A population-based study was undertaken to evaluate these differences for deaths in Connecticut in 1979 and 1980. Analyses for 1979 and 1980 were comparable. Of 28,440 deaths in 1980, 14% underwent autopsy. Differences by sex, race, marital status, age, underlying cause of death, and place of occurrence of death were demonstrated when decedents who underwent an autopsy were compared with all decedents. Non-white men who died at 20 to 29 years of age were the most likely group to undergo an autopsy. Proportions of autopsies, by major category of underlying cause, did not parallel the leading causes of death. Deaths due to trauma accounted for a larger proportion of all autopsies (25%) than of all deaths (6%). If a death certificate indicated a circulatory disorder or neoplasm as the cause of death, an autopsy was less likely to be performed. A disproportionately high number of autopsies were performed for deaths occurring in hospitals; the converse was true for deaths occurring in nursing-convalescent homes. We conclude that knowledge of the ways in which autopsies are biased will permit broader use of the sensitive and specific information that can be obtained from them.