It has been suggested that empiric broad-spectrum antibiotics, instituted for fever in the presence of granulocytopenia, should continue to be administered, even when infection is not demonstrable, to those patients who remain persistently febrile and granulocytopenic. Therefore, the consequences of discontinuing antibiotics when the presence of infection is doubted in this setting were evaluated. In 16 (3.7 percent) of 429 episodes of fever and granulocytopenia for which empiric antibiotic therapy was instituted, after approximately four days, persistence of both fever and granulocytopenia was found, and yet infection was prospectively classified at that time as "doubtful." The initial empiric antibiotic regimen was therefore discontinued after a mean of 4.8 (median 5.0) days. Discontinuation of antibiotics proved appropriate for half of the patients; eight patients received no systemic therapeutic antibiotics with no evidence of infection during a period of at least two weeks. The other eight patients had antibacterial antibiotics reinstituted within a mean of 2.4 days; six infections were subsequently demonstrable. Six of these eight patients also required or were believed to require antifungal therapy with intravenous amphotericin B for presumed fungal infections. Patients with relapsed leukemia or lymphoma and those with a likelihood of continued profound granulocytopenia (counts below 100/microliters) or both were the ones who tended to require reinstitution of antibiotics. Discontinuation of antibiotics when infection was considered doubtful despite persistence of both fever and granulocytopenia was, therefore, successful in eight of 16 patients. Reinstitution of antibiotics was required in the eight remaining patients. No definite rule appears to be applicable to all patients.