A three-arm prospective randomized trial was designed to compare the efficacies of piperacillin plus vancomycin, ceftazidime plus vancomycin, or all three drugs as initial therapy for fever in neutropenic cancer patients. The objectives were to determine whether a broad-spectrum penicillin was as effective as a broad-spectrum cephalosporin and whether two beta-lactam antibiotics were more effective than one. Four hundred and seventy of the 519 febrile episodes entered in the study could be evaluated for response. Ceftazidime plus vancomycin was significantly more effective than piperacillin plus vancomycin, considering all febrile episodes (79 percent versus 61 percent, p = 0.001), documented infections (79 percent versus 57 percent, p = 0.004), gram-negative infections (88 percent versus 47 percent, p = 0.001), and bacteremias (81 percent versus 51 percent, p = 0.01). The addition of piperacillin to ceftazidime (piperacillin plus ceftazidime and vancomycin versus ceftazidime plus vancomycin) did not improve the response rate and was associated with a significantly higher incidence of skin rash. Vancomycin plus ceftazidime provides adequate antibiotic coverage for initial treatment of fever in neutropenic patients. This combination was equally effective, less expensive, and less toxic than the double beta-lactam combination used in this study.