In a prospective cohort of 2,647 human immunodeficiency virus type 1 (HIV-1) seropositive homosexual men enrolled in Baltimore, Chicago, Los Angeles, and Pittsburgh, 891 developed clinical acquired immunodeficiency syndrome (AIDS) between June 1984 and January 1992. Cox proportional hazards models were used to examine temporal trends in survival after AIDS for specific diagnoses, controlling for level of immunosuppression at diagnosis, age, race, and geographic location. Median survival time following AIDS onset increased from 11.6 months in 1984-1985 to 19.5 months in 1988-1989; for those diagnosed in 1990-1991, the median survival time dropped to 17.2 months. Trends in improved survival were diagnosis-specific. Survival after Pneumocystis carinii pneumonia consistently improved from 1984 to 1991 (p < 0.001). Compared with men diagnosed in 1984-1985, those diagnosed with P. carinii pneumonia in 1990-1991 had one-tenth the hazard of dying. For men with > or = 100 helper T-lymphocytes (CD4+ cells) when diagnosed with Kaposi's sarcoma, the relative hazards (95% confidence intervals) of dying after Kaposi's sarcoma were 0.8 (0.42-1.60) in 1986-1987, 0.7 (0.34-1.58) in 1988-1989, and 0.6 (0.19-1.61) in 1990-1991 compared with those diagnosed before 1986. Men with < 100 CD4+ cells when diagnosed with Kaposi's sarcoma did not demonstrate a consistent change in their subsequent survival. After a nonsignificant (p > 0.05) initial improvement in prognosis, there has not been a significant improvement in survival for men who presented with other opportunistic infections. Observed increases in overall survival probably relate to improved treatment of patients who develop P. carinii pneumonia. Limited improvement in survival following other AIDS diagnoses indicates the need for developing effective treatment against these diseases.