The object of this study was to determine whether there were any differences between the 'typical' child with fever and neutropenia and their adult counterpart with regard to infection type and outcome, by analysis of 3080 patients, including 759 children < 18 years of age and 2321 adults. These represented patients randomized in previous trials, between 1986 and 1994, which compared empirical antibiotic regimens for fever in neutropenic patients. There were fewer childhood acute myeloid leukaemia patients than adults but more acute lymphoblastic leukaemia cases and more with solid tumours undergoing intensive myelosuppressive therapy. The children were less likely to be undergoing first induction therapy but the relative incidence of patients receiving relapse schedules or maintenance therapies were not significantly different in the two age groups. Children less frequently had a defined site of infection than adults and where they had a defined site there were more upper respiratory tract but fewer lung infections. There was a similar low incidence of shock at presentation in the two groups but the children's median neutrophil count was lower, and their median duration of granulocytopenia before the trial was shorter. The incidence of bacteraemia was similar, but clinically documented infection was less frequent and fever of unknown origin consequently more common in children. Children developed more streptococcal bacteraemias and fewer staphylococcal bacteraemias than adults (P=0.003) but the relative incidence of various gram-negative species was similar (P=0.57). In general, the children had a better overall success rate and lower mortality than adults. Death from infection was only 1% in children versus 4% in adults (P=0.001), and time to defervescence was shorter in children. In the younger age group, univariate logistic regression models showed high temperature, prolonged neutropenia before the trial and shock as prognostic indicators for the presence of bacteraemia. Solid tumour patients were significantly less likely to have a bacteraemia. Multivariate analysis confirmed the independent prognostic value of these indicators. Using the logistic equation of the selected model, the overall discriminant ability was poor. However, it was possible to identify a small subgroup without shock or high fever and with a short prior duration of neutropenia which carries a particularly low risk of bacteraemia, who could be considered for early discharge, monotherapy and shortened courses of antibodies, in prospective trials.