Following a 4-year controlled trial comparing early and later zidovudine treatment, we conducted an additional 3-year follow-up. Of the original 338 patients, 275 participated. Clinical outcome measures were AIDS and death. In the early therapy group (n = 170), 67 patients progressed to AIDS compared with 85 in the later therapy group (n = 168); the relative risk (RR) comparing early with later therapy was 0.72% (95% confidence interval [CI] 0.52-0.99; p = 0.044). The early therapy group had 74 deaths compared with 73 in the later therapy (RR = 0.98; 95% CI, 0.71-1.36; p = 0.91). The early group had a peak CD4+ count increase at 1-2 months and a delay of 1 year before CD4+ counts fell below baseline. For patients who received zidovudine for more than the median duration (20.3 months) before their first AIDS diagnosis, the RR for death was 2.08 (95% CI, 1.36-3.19, p = 0.001). Additional factors independently associated with poor prognosis following AIDS were a CD4+ count of < 100 cells/mm3 and increased severity of the first AIDS diagnosis, whereas use of another antiretroviral agent was associated with improved survival. We conclude that early zidovudine therapy delays progression to AIDS but does not affect survival. Patients who progress to AIDS while on prolonged zidovudine monotherapy many benefit from a change to other antiretroviral therapy(ies).