From 1986 to 1992, the Febrile Aplasia Study Group conducted a series of studies involving severely neutropenic patients. The average duration of neutropenia was 21 days, following chemotherapy for leukaemia, or chemotherapy/radiotherapy as part of a conditioning regimen for autologous or allogeneic bone marrow transplantation. A total of 591 evaluable febrile episodes were randomized to treatment with either ceftazidime 3 g daily + amikacin (the reference regimen; n=246), ceftazidime alone (n=98), ceftazidime + vancomycin (n=77), ceftazidime + ciprofloxacin (n=64) or piperacillin/tazobactam + amikacin (n=106). Only three patients treated with the reference dose of ceftazidime died or suffered serious morbidity from infections caused by Gram-negative bacteria. Piperacillin/tazobactam + amikacin was the only antibiotic regimen to have an effect significantly different from the reference regimen. Piperacillin/tazobactam + amikacin produced a higher rate of defervescence at 72 h (P=0.003), fewer days of fever (P < 0.001), fewer superinfections (P=0.018), a less frequent requirement for addition of vancomycin (P=0.01) and a higher incidence of treatment judged to be a 'complete success' (enduring defervescence without a change in antibiotics) (P=0.04). Despite the improved control of Gram-positive microorganisms, the infection-related death rate remained unchanged from 1987 to 1992. An increase in disseminated aspergillosis compensated for the reduction in lethal Gram-positive septicaemia.