We prospectively compared sputum induction with bronchoalveolar lavage (BAL) in HIV positive patients presenting with acute respiratory episodes and also assessed the effects of using an experienced respiratory physiotherapist on the diagnostic yield from induced sputum. One hundred and fifty-one consecutive patients underwent sputum induction, in 96 the procedure was supervised by nursing and medical staff with no specific expertise (group 1); in 55 patients a physiotherapist supervised sputum induction (group 2). Nine patients refused BAL having undergone sputum induction. Of the remaining 142 patients sputum induction failed (no sample expectorated) in 28 patients (25 from group 1 and three from group 2), the sample was inadequate (the material expectorated was not from the lower respiratory tract) in 29, and was adequate in 85 patients. Pneumocystis carinii was diagnosed in 82 patients (51 from group 1 and 31 from group 2). The sensitivity of induced sputum for the diagnosis of P. carinii was 13 % and of BAL was 77 %. In the subgroup of patients with an adequate induced sputum sample, the sensitivity of induced sputum was 28 % and of BAL was 73 %. Of the remaining 60 patients, 27 had other diagnoses made by induced sputum and BAL (eight patients), BAL only (15 patients) and induced sputum only (four patients). Eleven patients had bronchitis and responded to oral antibiotics. In 22 patients induced sputum and BAL were negative; alternative diagnoses were established by lung biopsy or by culture of blood, urine or CSF. During sputum induction, 15 patients had nausea and vomiting, eight became dyspnoeic, three had intractable cough and one developed acute bronchoconstriction; 17 patients found the procedure unpleasant. Compared with BAL, induced sputum has a lower diagnostic yield for P. carinii and other pathogens. Use of experienced, dedicated personnel increases the number of successful attempts at sputum induction but does not increase the diagnostic yield. Fibreoptic bronchoscopy and bronchoalveolar lavage remain necessary for patients with negative results from induced sputum and those whose disease course is at variance with the diagnosis made by sputum induction. © 1991 The British Society for the Study of Infection.