Many patients infected with the human immunodeficiency virus (HIV) with symptoms suggestive of pneumonia are treated empirically for Pneumocystis carinii pneumonia (PCP), although other bacterial infections (e.g., tuberculosis) and pulmonary Kaposi's sarcoma may cause identical symptoms. Empiric treatment for PCP may result in misdiagnosis and mistreatment. When the outcomes of cytologically confirmed versus empirically treated PCP cases were evaluated, the most important predictors of in-hospital mortality were severity of illness and use of bronchoscopy. Persons who did not undergo bronchoscopy had higher mortality rates than patients negative by bronchoscopy or cytologically confirmed as positive for PCP (22% vs. 11% vs. 14%, P < .01), although severity of illness and timing of anti-PCP medications did not differ significantly. Compared with cytologically confirmed cases, persons who did not have bronchoscopy were more likely to die than were bronchoscopy-negative patients (P < .05), after adjusting for severity of illness. Bronchoscopy use may have contributed to better outcomes for persons treated for HIV-related PCP.