The use of vancomycin as part of the initial antibiotic therapy of febrile neutropenic patients has become a controversial issue. Some studies support its incorporation in the initial regimen, and others suggest that vancomycin can be added later. We examined this issue in a prospective, randomized trial. We randomized 127 febrile neutropenic patients to receive either ceftazidime alone or ceftazidime plus vancomycin as the initial empiric antibiotic treatment. We added vancomycin to the ceftazidime arm of the study when fever persisted after 96 h of monotherapy, when new fever occurred after this time, or when a moderately ceftazidime-resistant gram-positive bacterium was isolated. Each of these regimens had similar initial response rates, similar durations of initial fever, similar frequencies of new fever during therapy, similar microbiological cure rates, similar superinfection rates, and similar survival rates. We observed more renal and cutaneous toxicities in patients receiving vancomycin and ceftazidime as initial therapy. We conclude that ceftazidime is appropriate as initial therapy for febrile neutropenic patients and that the addition of vancomycin is appropriate when fever persists after 4 days of monotherapy or when fever recurs following an initial response.