Febrile episodes in neutropenic patients with cancer should be treated empirically with broad-spectrum antibiotics that are bactericidal and potentially synergistic against the presumed pathogens. At present, the combination of carbenicillin or ticarcillin and amikacin seems optimal. In patients whose fever responds to therapy but who remain neutropenic, the initial empiric regimen should be continued as long as neutropenia persists. The risk of superinfection associated with prolonged antimicrobial therapy in neutropenic patients should be investigated further. When a probable pathogen is isolated from an unresponsive patient, antimicrobial therapy should be adjusted in light of the in vitro sensitivity of the infecting microorganisms, the possibility of localized infection should be carefully investigated, and granulocyte transfusions should be considered. However, empiric administration of granulocytes is probably not indicated. In the group of neutropenic patients that is most difficult — those who have fever but whose infections cannot be microbiologically documented and who fail to respond to empiric antimicrobial therapy — the early administration of amphotericin B should be seriously considered, although there is no evidence that the drug modifies greatly the overall outcome in these patients. Whether granulocyte transfusions should be combined with administration of amphotericin B in these patients remains to be investigated further.