An empirical approach to treating Pneumocystis carinii pneumonia was adopted in a prospective study of 73 men with antibodies to human immunodeficiency virus 1 (HIV-1) presenting with respiratory problems. At presentation 49 patients (group 1) were thought to have a history, findings at clinical examination, chest radiography, and arterial blood gas tensions typical of pneumocystis pneumonia, and empirical treatment was begun immediately. Twenty four patients (group 2) were thought to have features not typical of pneumocystis pneumonia. All patients were subsequently referred for bronchoscopy to determine the diagnosis. In group 1 four patients were excluded from the analysis because bronchoscopy was not possible. Of the remaining 45, 42 had pneumocystis pneumonia, which was diagnosed at bronchoscopy in 40, and on the basis of the clinical response to co-trimoxazole in two who had negative results from investigations. Of the three patients without pneumocystis pneumonia, one patient with lymphoid interstitial pneumonitis and Branhamella catarrhalis infection would have failed to respond to empirical treatment. The other two had multiple bacterial pathogens at bronchoscopy; one already had Kaposi's sarcoma and the other would have been misdiagnosed as having AIDS. In group 2 a specific diagnosis was made at bronchoscopy in 21 cases, including pneumocystis pneumonia in seven (all had atypical chest radiographs). In three cases no diagnosis was made and spontaneous recovery occurred. Adopting an empirical approach to treatment for typical pneumocystis pneumonia (group 1) led to the correct treatment in 43 of 45 cases (95%) and would have saved 44 of the 45 of bronchoscopies in this group. Adopting an empirical approach would have caused one patient to be misdiagnosed as having AIDS. Overall, 44 out of 69 bronchoscopies (64%) would have been saved; the specificity for the diagnosis of pneumocystis pneumonia was 85% and the sensitivity was 85%. Adopting an 'empirical' treatment policy for typical pneumocystis pneumonia will cause a large reduction in the number of 'high risk' bronchscopies performed.