Of 545 patients expected to develop prolonged neutropenia and randomized to received trimethoprim-sulfamethoxazole (TMP-SMZ) or placebo, 342 were evaluable for occurrence of infection or bacteremia. Some centers used oral nonabsorbable antibiotics in addition. Infection occurred in 64 (39%) of 165 placebo recipients and 46 (26%) of 177 TMP-SMZ recipients (P = .016), whereas bacteremia occurred in 32 (19%) and 22 (12%), respectively (P = .106, difference not significant [NS]). In the 139 patients with acute nonlymphocytic leukemia (ANLL), infection occurred in 35 (55%) of 64 placebotreated patients and 31 (41%) of 75 TMP-SMZ-treated patients (P = .162, NS), whereas bacteremia occurred in 15 (23%) and 18 (24%; NS), respectively. Excluding patients with ANLL, infection occurred in 29 (29%) of 101 placebo-treated patients and 15 (15%) of 102TMP-SMZ recipients (P = .038), whereas bacteremia occurred in 17 (17%) and four (4%; P = .005), respectively. Gram-positive cocci were isolated less frequently from TMP-SMZ-treated, bacteremic patients, but more of their isolates were resistant to TMP-SMZ than were those from placebo recipients.