Two antibiotic regimens, ceftazidime plus amikacin and ceftazidime plus vancomycin, were compared in a prospective, randomized clinical trial as empiric therapy in febrile granulocytopenic children with cancer. The rate of response was similar in the two groups (66% vs. 77%). The prevalence of secondary gram-negative bacteremia was higher--but not significantly higher--in the group receiving vancomycin. Adverse reactions also occurred more often in the latter group (35% vs. 4%). Mortality did not differ significantly in the two groups. Adjustment for independent predictors of response to treatment by means of multivariate analysis confirmed the lack of any remarkable difference between the responses to the two regimens. We conclude that the use of vancomycin instead of amikacin in combination with ceftazidime does not significantly improve the outcome of treatment of fever and infection in granulocytopenic children with cancer and that the use of vancomycin is associated with an increased frequency of both secondary infections due to gram-negative bacteria and adverse reactions.