The utility of bronchoalveolar lavage (BAL) in determining the causative agent of pulmonary infiltrates in patients with acute leukemia is not known. We retrospectively evaluated the diagnostic yield of BAL in 22 adults with acute leukemia and compared the results with those at autopsy performed within three weeks of BAL. All patients had neutropenia and thrombocytopenia at the time of BAL, were receiving broad-spectrum antibacterial agents, and 15 were also receiving amphotericin B before BAL. The median interval between the detection of pulmonary infiltrates and BAL was seven days (range, 0 to 23 days); the median interval between BAL and autopsy was nine days (range, 1 to 20 days). The diagnostic yield of BAL was 15 percent (3 of 20 specific diseases); all three were Candida pneumonia. The sensitivity of BAL was 75 percent and its specificity 100 percent, for Candida pneumonia. BAL did not result in a specific diagnosis for the 17 remaining diseases, nine of which were Aspergillus pneumonia. In seven patients in whom autopsy was performed within 72 hours of BAL, lavage results correlated with those of autopsy in only one who had Candida pneumonia. All BAL cultures were falsely positive, except in four cases of Candida pneumonia. The therapeutic regimen was not modified according to the BAL results in any of the 22 patients. There were no major complications associated with the procedure.