OBJECTIVE To determine the association between elapsed time since starting zidovudine and survival in patients with acquired immunodeficiency syndrome (AIDS). METHODS Inception cohort and observational study of patients treated and not treated with zidovudine. METHODS Fifty-one centers in 17 European countries. METHODS A total of 4484 patients diagnosed as having AIDS from 1979 to 1989 who survived their initial AIDS-defining event and who had not started zidovudine before AIDS diagnosis. METHODS Use of zidovudine and mortality. RESULTS Among patients who did not receive zidovudine, the death rate was approximately constant for the first 5 years after AIDS diagnosis. For patients treated with zidovudine, the death rate within the first year since starting zidovudine was markedly lower than for untreated patients who had developed AIDS at the same time (relative rate, 0.47; 95% confidence interval [CI], 0.42 to 0.51). For longer times since starting zidovudine, the association with reduced mortality rate was diminished, and for patients surviving more than 2 years since starting zidovudine, the death rate was greater than for untreated patients who had developed AIDS at the same time (relative rate, 1.35; 95% CI, 1.15 to 1.58). Adjustment for other prognostic factors failed to substantially affect this observation. CONCLUSIONS When initiated after the time of AIDS diagnosis, zidovudine was associated with improved prognosis but for no more than 2 years after starting therapy.