In a cohort of 5833 subjects in whom the acquired immunodeficiency syndrome (AIDS) was diagnosed in New York City before 1986, the cumulative probability of survival (mean +/- SE) was 48.8 +/- 0.7 percent at one year and 15.2 +/- 1.8 percent at five years. The group with the most favorable survival rate--white homosexual men 30 to 34 years old who presented with Kaposi's sarcoma only--had a one-year cumulative probability of survival of 80.5 percent; that group was used as the reference group in assessing the effect of five variables: sex, race or ethnic background, age, probable route of acquiring AIDS (risk group), and manifestations of AIDS at diagnosis. The range in the mortality rate was greater than threefold, depending on these variables. Black women who acquired the disease through intravenous drug abuse, for example, had a particularly poor prognosis. The manifestations of disease at diagnosis had the most influence on survival, accounting on average for 56.3 percent of the excess risk. This variable was followed in importance by age (12.2 percent), race or ethnicity (10.6 percent), risk group (8.4 percent), and sex (8.0 percent), with 4.5 percent of the risk attributable to interactions between variables. When we compared subcohorts based on the year of diagnosis (1981 through 1985), we found a significant improvement in the one-year cumulative probability of survival among subjects with Pneumocystis carinii pneumonia, but not among subjects without P. carinii pneumonia.