Objective To describe the natural history of advanced HIV disease in patients treated with zidovudine. Design Longitudinal, observational study. Setting Twelve academic and community-based sites. Patients, participants Eight hundred and sixty-three patients with AIDS or AIDS-related complex (ARC) with a CD4+ lymphocyte count 6/1, who first received zidovudine between 15 April 1987 and 14 April 1988. Main outcome measures Survival, progression to AIDS and first development of specific opportunistic illness. Results Median survival after initiation of zidovudine therapy ranged from >900 days in patients with a baseline CD4+ lymphocyte count ≥150 × 106/1 to 560 days in patients with a CD4+ lymphocyte count < 50 × 106/1. Other factors associated significantly with poorer survival were diagnosis of AIDS (versus ARC), baseline age ≥40 years, hematocrit < 35%, and diminished functional status. In patients with ARC at enrollment, median time of progression to AIDS ranged from 810 days in patients with a CD4+ lymphocyte count ≥150 × 106/1 to 310 days in patients with a CD4 + lymphocyte count < 50 × 106/1. Rates of development of specific opportunistic infections or neoplasms and HIV encephalopathy were determined for different baseline CD4 + lymphocyte count ranges. Myelosuppression was significantly more common in patients with CD4+ lymphocyte counts ≥100 × 106/1. Sixty-five per cent of patients with a CD4 + lymphocyte count ≥100 × 106/1 and 51% with a CD4+ lymphocyte count < 100 × 106/1 continued to receive zidovudine 2 years after starting therapy. Conclusions We describe the natural history of a cohort of patients treated with zidovudine for advanced HIV disease. These CD4+ lymphocyte count-stratified estimates of disease progression should provide prognostic information useful in the clinical management of advanced disease and the design of future studies.