OBJECTIVE To test the hypothesis that the therapeutic success rate of oral atovaquone is not worse than that of intravenous pentamidine in the primary treatment of mild and moderate Pneumocystis carinii pneumonia in patients with the acquired immunodeficiency syndrome and to detect differences in the toxicity rates of the two treatments. METHODS Patients were randomly assigned to receive 21 days of open-label therapy with either atovaquone, 750 mg orally with meals three times daily, or intravenous pentamidine, 3 to 4 mg per kg body weight once daily. METHODS Multicenter study including university and community treatment facilities. METHODS Patients with human immunodeficiency virus infection and clinical presentations consistent with mild or moderate P. carinii pneumonia were eligible. For efficacy and safety analyses, patients with histologically confirmed P. carinii pneumonia were emphasized. METHODS Patients were monitored by clinical and laboratory evaluations for therapeutic efficacy and adverse events during the acute treatment phase and for 8 weeks after therapy was discontinued. RESULTS As initial therapy for a histologically confirmed episode of P. carinii pneumonia, 56 patients received atovaquone and 53 received pentamidine. More patients were successfully treated with atovaquone (57%) than with pentamidine (40%), a difference of 17% (95% CI, -3% to 38%; P = 0.085), but more patients failed to respond to atovaquone (29%) than to pentamidine (17%), a difference of 12% (CI, -6% to 29%; P = 0.18). Discontinuation of original therapy because of treatment-limiting adverse events was more frequent in the pentamidine group (36%) than in the atovaquone group (4%) (difference, -32%; CI, -48% to -17%; P < 0.001). Nine patients in each treatment group died during the study. CONCLUSIONS Oral atovaquone and intravenous pentamidine have similar rates for successful treatment of mild and moderate P. carinii pneumonia, but atovaquone has significantly fewer treatment-limiting adverse events.