Adjunctive corticosteroid therapy is recommended for selected human immunodeficiency virus (HIV)-infected patients with presumed Pneumocystis carinii pneumonia. Because corticosteroids may exacerbate undiagnosed tuberculosis, we evaluated the frequency with which tuberculosis in HIV-infected patients mimics P carinii pneumonia. Over a 12-month period, we identified 105 HIV-infected patients with pleuropulmonary tuberculosis and 84 patients with P carinii pneumonia who were sufficiently hypoxemic to warrant corticosteroid therapy. Of the 105 patients with tuberculosis, acid-fast smears of clinical samples were positive in 49 cases, and chest roentgenographic findings suggested tuberculosis in an additional 44 cases. The 12 patients with negative acid-fast smears and nonspecific chest roentgenographic findings presented a potential diagnostic dilemma between tuberculosis and P carinii pneumonia. Pneumocystis carinii pneumonia should not have been a presumptive diagnosis of eight of these 12 patients because of absence of pulmonary symptoms and chest roentgenographic abnormalities (four cases), a CD4 count greater than 500/cu mm (three cases), or marked lymphadenopathy suggestive of tuberculosis (one case). Thus, only 4 percent (4/105) of HIV-infected patients with pleuropulmonary tuberculosis had clinical and chest roentgenographic features mimicking P carinii pneumonia. Two of these four patients were sufficiently hypoxemic to warrant corticosteroid therapy. Thus, if corticosteroids had been routinely used during the study period, 84 patients with P carinii pneumonia would have been treated, including two patients with undiagnosed tuberculosis. We conclude that the use of corticosteroids for presumed P carinii pneumonia carries a small but acceptable risk of inadvertent exacerbation of tuberculosis, provided clinical and chest roentgenographic features do not suggest tuberculosis.