We assessed two antibiotic regimens--vancomycin, ticarcillin, and amikacin, as compared with a vancomycin placebo, ticarcillin-clavulanate, and amikacin--as initial empirical therapy for febrile, neutropenic children with cancer. In a randomized, double-blind clinical trial, the planned 10-day treatment was unsuccessful in 15 percent of the vancomycin, ticarcillin, and amikacin group (n = 53), as compared with 38 percent of the group receiving ticarcillin-clavulanate and amikacin (n = 48) (P = 0.010). Of 10 episodes of breakthrough bacteremia, 9 (1 fatal) occurred in patients treated with ticarcillin-clavulanate and amikacin (P = 0.006). Each of the 10 microbial isolates was a gram-positive bacterium with similar susceptibilities to vancomycin and ticarcillin-clavulanate in vitro. Both regimens were well tolerated. None of the patients had detectable renal dysfunction, but those receiving vancomycin, ticarcillin, and amikacin were more likely to have twofold increases in serum hepatic-enzyme activity. Rashes consistent with the "red-man" syndrome occurred in three patients upon the infusion of vancomycin and in three others who received a placebo. We conclude that the combination of vancomycin, ticarcillin, and amikacin is more effective than ticarcillin-clavulanate and amikacin as empirical antibiotic therapy in clinical settings in which gram-positive bacteremias are a serious problem.