The purpose of this study was to assess the sensitivity and specificity of the polymerase chain reaction (PCR) on induced sputum (IS) for the diagnosis of Pneumocystis carinii pneumonia (PCP) in HIV-infected patients, as well as its diagnostic value and cost as a routine clinical tool. Forty-nine patients with suspected PCP who had IS were studied and if negative, followed by bronchoalveolar lavage (BAL). Pneumocystis carinii was detected in these samples using standard staining techniques. Polymerase chain reaction was used with IS samples in a blinded fashion. The patients with negative BAL samples were closely monitored for 1 month. In the absence of any clinical or radiologic features of PCP during this period, they were considered as being free of PCP. The cost analysis considered only the direct costs of the various tests in three diagnostic strategies: routine BAL (BAL); IS with standard staining, if negative, followed by BAL (IS); and IS with standard staining followed, if negative, by PCR on IS samples (PCR-IS). Using standard staining P carinii was found in 13 cases (6 IS and 7 BAL). None of the 36 patients with negative BAL developed further signs of PCP. Thus, the prevalence of PCP was 26.5% and the sensitivity and specificity of BAL were 100%. Standard staining of IS had a specificity of 100% and a sensitivity of 46.5%. The sensitivity and specificity of PCR-IS were each 100%. The costs of strategies BAL, IS, and PCR-IS were $14,010, $18,300, and $18,040, respectively. The costs of the BAL strategy depended only on the cost of the relevant tests, whereas the costs of strategies IS and PCR-IS depended on the costs of the tests, the sensitivity of IS with standard staining, and the prevalence of PCP in the test population. The routine clinical use of PCR-IS is currently limited by the time required to obtain the results.