We reviewed 60 consecutive flexible bronchoscopies done during a 36-month period in 48 pediatric cancer patients with undiagnosed pulmonary infiltrates. Diagnostic procedures during bronchoscopy included 40 brushings, 50 bronchoalveolar lavages, and 6 transbronchial and mucosal biopsies. A total of 16 specific diagnoses were made by bronchoscopy (27% diagnostic yield), including infection (12), pulmonary leukemia (3), and lymphoma (1). The largest proportion of specific diagnoses came from lavage (14/50) and the smallest from brushings (1/40). Biopsies were also useful for selected patients. The low overall yield for bronchoscopy was probably due to the routine use of empiric broad-spectrum antibiotics and antifungal therapy, as well as trimethoprim-sulfamethoxazole prophylaxis for Pneumocystis carinii pneumonitis. Subsequent specific diagnoses were obtained by other procedures (open biopsy, needle aspiration, or autopsy) for 10 patients with negative bronchoscopy results and 3 patients with diagnostic bronchoscopies. These additional diagnoses included 7 infections (Pneumocystis carinii (1), Candida tropicalis (1), cytomegalovirus (1), and Aspergillus (4), and 6 other diagnoses with nonspecific histologic findings. A positive bronchoscopy result may be useful, but negative bronchoscopy findings do not justify delaying other diagnostic procedures or discontinuing antibiotic and antifungal therapy in children with cancer and pulmonary infiltrates.