Several definitions of response have been used to classify the results of empiric antibiotic therapy in febrile neutropenic patients, but none has become universally accepted. In two consecutive Canadian multicentre randomized trials, outcomes were classified in a scheme which allowed us to assess the effect of different outcome definitions on interpretation of the results of the studies. We found that the use of different outcome definitions resulted not only in variations in absolute response rates, but also affected comparisons among patient groups. Thus, both the adverse prognostic effect of leukemia as an underlying disease and the poorer results obtained using aztreonam-containing regimens were evident only with certain outcome definitions. These results indicate a need to pay careful attention to outcome definitions in interpreting the results of studies of antibiotic therapy in febrile neutropenic patients.