A randomized, unblinded study compared aerosolized pentamidine, 300 mg every month, and dapsone, 50 mg/day, for secondary prophylaxis of pneumocystosis in 196 AIDS patients. The study was prematurely discontinued due to excess mortality in the dapsone group. After a mean follow-up of 13 +/- 6.4 months, 22 (21%) of 103 patients in the pentamidine group were dead compared with 39 (42%) of 93 receiving dapsone; the estimated mortality rates at 18 months were 24.6% and 53.1%, respectively (P < .003, log-rank test). A negative interaction was observed between zidovudine and dapsone (P < .049, interaction test of Cox model), and the mean CD4 cell count during the study was lower in the dapsone (49 +/- 61/mm3) than in the pentamidine group (83 +/- 88/mm3; P < .002, t test). The lower survival might also be related to the oxidative effect of dapsone or to the addition of iron protoxalate to dapsone in this study. These results suggest caution in using dapsone as long-term therapy in advanced human immunodeficiency virus infection, especially in those receiving zidovudine.