The results of therapy with carbenicillin plus trimethoprim-sulfamethoxazole (C-T/S) were compared to those obtained with carbenicillin plus gentamicin (C-G) in a prospective double-blind study of empiric antibiotic therapy in granulocytopenic patients. Patients were stratified into two groups: favorable-prognosis, group 1 (carcinoma, lymphoma, multiple myeloma), or unfavorable-prognosis, group 2 (acute leukemia, bone marrow transplantation), based on anticipated duration of granulocytopenia. Over-all, empiric antibiotic trials were more often successful (P = 0.004) in group 1 (55 of 62 patients or 89 per cent) than in group 2 (42 of 64 patients, 66 per cent)mwithin group 1, there was a favorable outcome in 30 of 32 (94 per cent) C-T/S trials and in 25 of 30 (83 per cent) C-G trials (P = 0.25); within group 2, there was a favorable outcome in 23 of 30 (77 per cent) C-T/S trials and in 19 of 34 (56 per cent) C-G trials (P = 0.14), Combined results in both groups indicated a higher proportion of favorable outcome in C-T/S trials (53 of 62, 85 per cent) than in C-G trials (44 of 64, 69 per cent). Further analysis (Manetl-Naenszel test) showed the over-all difference in outcome to be significant (P = 0.049), but the general applicability of this result may be limited by the rather low incidence of gram-negative bacterial infections in this study. There was no difference between the treatment regimens in antibiotic toxicity, and serious superinfection occurred only in group 2 patients (21 per cent of trials), equally divided between treatment arms. Initial protocol dosing achieved target plasma levels of trimethoprim (3 to 8 micrograms/ml) or gentamicin (4 to 10 micrograms/ml) in 57 of 68 (84 per cent) C-T/S trials compared to 21 of 60 (35 per cent) C-G trials.