In this multicenter study (ISS 902), 554 previously untreated patients with <500 CD4 cells/mm3 and mildly symptomatic human immunodeficiency virus disease were randomized to receive zidovudine or didanosine (ddI). After a mean follow-up of 20 months, 80 patients (40 zidovudine, 40 ddI) had died and 146 had at least one AIDS-defining event (73 zidovudine, 73 ddI). Overall, no difference was found between treatments with respect to progression to AIDS or death. The analysis of relative risk (RR) of progression over time, however, showed an initially minor risk for zidovudine patients and an inversion in the zidovudine-ddI RR in the second and third years of follow-up. Didanosine showed a greater effect on CD4 cell count response. The two drugs confirmed the toxicity patterns already reported in other trials, with a low occurrence of pancreatitis (ddI 1.3%, zidovudine 0.4%). The overall results suggest that, in this population, zidovudine and ddI monotherapies have comparable long-term clinical efficacy and that more powerful regimens should be preferred.