BACKGROUND We evaluated the effectiveness of three treatment strategies for the prevention of a first episode of Pneumocystis carinii pneumonia in patients infected with the human immunodeficiency virus (HIV). METHODS In an open-label trial, 843 patients with HIV infection and fewer than 200 CD4+ cells per cubic millimeter received zidovudine plus one of three randomly assigned prophylactic agents, beginning with trimethoprim-sulfamethoxazole, dapsone, or aerosolized pentamidine and followed by a defined sequence of other drugs to be used in cases of intolerance. RESULTS The estimated 36-month cumulative risks of P. carinii pneumonia were 18 percent, 17 percent, and 21 percent in the trimethoprim-sulfamethoxazole, dapsone, and aerosolized-pentamidine groups, respectively (P = 0.22). The difference in risk among treatment strategies was negligible in patients entering the study with 100 or more CD4+ lymphocytes per cubic millimeter. In those entering with fewer than 100 CD4+ cells per cubic millimeter, the risk was 33 percent with aerosolized pentamidine, as compared with 19 percent with trimethoprim-sulfamethoxazole and 22 percent with dapsone (P = 0.04). The lowest failure rates occurred in patients receiving trimethoprim-sulfamethoxazole, and failures were more common with 50 mg of dapsone than with 100 mg. Toxoplasmosis developed in less than 3 percent of patients. Of the patients assigned to the two systemic therapies, only 23 percent were receiving their assigned drug and dose when they completed the study. The median survival was approximately 39 months in all three groups, and the mortality attributable to P. carinii pneumonia was only 1 percent. CONCLUSIONS In patients with advanced HIV infection, the three treatment strategies we examined have similar effectiveness in preventing P. carinii pneumonia. Strategies that start with trimethoprim-sulfamethoxazole or with high-dose dapsone, rather than aerosolized pentamidine, are superior in patients with fewer than 100 CD4+ lymphocytes per cubic millimeter.