In order to define mortality rates and clinical findings with prognostic value in febrile infections among our adult patients with acute leukemia we prospectively studied--during a 34 months period--177 episodes of clinical suspected infection which occurred in 49 patients. By means of univariate analysis and a subsequent multiple logistic regression study, the association between 27 clinical and microbiological data and febrile episode survival rates were evaluated. Both the overall mortality rate and the specific one for febrile episodes were high (44.9% and 12.7% respectively). An age over 30 years old (p = 0.025), fever lasting more than five days (p = 0.025), lung involvement (p = 0.001) and fungal isolation in a culture specimen (p = 0.005) were all associated with a higher episode mortality. However, only an age older than 30 years (adjusted odds ratio, A.O.R. = 1.118; 95% confidence interval, C.I.95% = 1.015-1.232; p = 0.025) and pneumonia (A.O.R. = 1.454; C.I.95% = 1.288-1.642; p < 0.001) remained as independent predictors of a greater mortality in the multivariate analysis. Although fever of unknown origin was associated with a better prognosis (p = 0.024) two other variables--viral infections (A.O.R. = 0.642; C.I.95% = 0.421-0.979; P = 0.041) and the isolation of two or more etiologic agents (A.O.R. = 0.795; C.I.95% = 0.651-0.972; p = 0.027)--had a protective value with the multiple regression analysis.