The authors examined the survival experience of 289 human immunodeficiency virus (HIV)-infected men to identify factors independently associated with survival in patients with very low CD4 counts (< or = 100/mm3). All subjects were HIV-infected men cared for at the San Francisco Veterans' Affairs Medical Center between January 1988 and November 1991. Survival was measured from the date on which a patient was first known to have a CD4 count < or = 100/mm3 until death or censoring. Factors potentially associated with survival were examined initially using the product limit (Kaplan-Meier) method; a multivariate model of survival was constructed using a proportional hazards (Cox) regression. Four variables were identified as independently associated with survival (p < 0.05) in the Cox proportional hazards model: CD4 count, hematocrit, azidothymidine use, and clinical stage (prior history of acquired immunodeficiency syndrome vs. no prior history). All 16 possible combinations of these four (dichotomized) variables were examined; eight different patterns of survival were detected. Identification of survival patterns that can be described by data obtained as part of routine clinical care has implications for patient care, the design of clinical trials, the study of mechanisms of progression of HIV-related immunosuppression, and planning of health care resource needs for populations of patients with very low CD4 counts.