BACKGROUND We conducted a trial to compare treatment with zidovudine or didanosine in patients with advanced human immunodeficiency virus type 1 (HIV-1) infection who had received little or no previous therapy with zidovudine. METHODS Six hundred seventeen patients with acquired immunodeficiency syndrome (AIDS), advanced AIDS-related complex (CD4 cell count, < or = 0.30 x 10(9)/L [300/microL]), or asymptomatic HIV (CD4 cell count, < or = 0.20 x 10(9)/L) received zidovudine, 500 mg/d of didanosine, or 750 mg/d of didanosine in a randomized, double-blind allocation, with cross-over to alternative medication after development of an end point or serious toxic effect. To be eligible, patients must have received either no or up to 16 weeks of zidovudine therapy before entry into the study. Primary end points were development of a new AIDS-defining event or death. Secondary clinical end points were new or recurrent AIDS-defining events, or death, and survival. RESULTS In the study as a whole, there were no differences in the relative risks (RRs) of the development of end points between treatment groups. However, there was a strong interaction between the relative efficacies of zidovudine and didanosine and previous experience with zidovudine. Among 380 patients with no previous zidovudine therapy, zidovudine was more effective than 750 mg/d of didanosine (RR, 1.43; 90% confidence interval [CI], 1.02 to 2.00), with a similar trend for zidovudine compared with 500 mg/d of didanosine (RR, 1.21; 90% CI, 0.86 to 1.71). However, among 118 patients with more than 8 weeks but no more than 16 weeks of previous zidovudine therapy, 500 mg/d of didanosine was more effective than zidovudine (RR, 0.48; 90% CI, 0.27 to 0.86); there was a similar trend for increased effectiveness of 750 mg/d of didanosine compared with zidovudine (RR, 0.61; 90% CI, 0.36 to 1.03). Among 119 patients who had some but no more than 8 weeks of previous zidovudine therapy, there were no significant differences among the treatment arms. Similar findings were noted in the analysis of the two secondary clinical end points. No significant differences were found in efficacy between the groups receiving 500 and 750 mg/d of didanosine. The major toxic effect associated with zidovudine was hematopoietic (granulocytopenia) and that associated with didanosine was pancreatitis (dosage, 750 mg/d). CONCLUSIONS In patients with advanced HIV disease, zidovudine appears to be more effective than didanosine as initial therapy; however, some patients with advanced HIV disease may benefit from a change to didanosine therapy after as little as 8 to 16 weeks of therapy with zidovudine.