We illustrate an analysis with classification and regression trees applied to survival data. Through this application, we provide a description of the opportunistic diseases and sociodemographic factors that contribute to survival among people with human immunodeficiency virus disease. The analyses are based on 43,795 cases reported to the Centers for Disease Control between January 1, 1984, and December 31, 1987. We used vital status as of December 31, 1989, to estimate mortality rates. We identified Kaposi's sarcoma and opportunistic diseases causing central nervous system damage (cryptococcosis, primary lymphoma of the brain, cytomegalovirus disease, and progressive multifocal leukoencephalopathy) as important predictors of death. In addition, advanced age at diagnosis (50+), race (white/other), and history of illicit drug use were found to be important determinants. Estimates of the cumulative probability of survival for subgroups of individuals defined by the tree structure illustrate the effect of these determinants on mortality. For the purpose of comparison, two proportional hazards models were also fit to the data using factors identified in the tree structure as the determinants of interest. This application illustrates the utility and limitations of both this new technique and proportional hazards models for epidemiologic research.