Although PCR detection of Pneumocystis carinii DNA has been described, little is known about the sensitivity or specificity of the assay in routine laboratory practice. We had the unique opportunity to use a bronchoalveolar lavage (BAL) specimen bank with samples for which the direct examination results for P. carinii were known. DNA purified from 129 selected specimens was amplified by using the primers described previously (A. E. Wakefield, F. J. Pixley, S. Banerji, K. Sinclair, R. F. Miller, E. R. Moton, and J. M. Hopkin, Mol. Biochem. Parasitol. 43:69-76, 1990). Of the 129 specimens, 37 were positive for P. carinii by direct examination. All 37 specimens were positive for P. carinii by PCR, yielding a 100% sensitivity and 100% negative predictive value for the assay. An additional 23 specimens were repeatedly positive for P. carinii by PCR but were not positive by direct examination. Review of the patient charts for these specimens with discordant results demonstrated that five of the patients were actually positive for P. carinii, as determined by either biopsy or examination of repeat or prior BAL specimens. A response to empiric therapy for P. carinii pneumonia was seen in an additional two patients. Of the remaining specimens, 8 produced no significant isolates other than P. carinii, while 12 contained culture-confirmed significant respiratory pathogens in addition to P. carinii (two fungal, nine bacterial, and one viral pathogen). Cytomegalovirus, which was of unknown significance, was isolated from 16 additional specimens. Overall, the specificity of the PCR assay was 79.3% compared to the results of direct examination. We hypothesized that the apparently poor specificity of the PCR assay was due to the increased sensitivity of the assay compared to that of direct examination. The sensitivity of the PCR assay was therefore assessed with BAL specimens containing P. carinii cysts. Serial dilutions of this preparation were evaluated by direct examination and PCR. PCR was found to be 100-fold more sensitive than direct examination, which detected one to two cysts per amplification. No false-positive results were detected in controls containing no DNA or by using target DNA from various fungal, viral, or bacterial respiratory pathogens. We conclude that PCR detection of P. carinii in BAL specimens is very sensitive and should be considered for patients whose specimens do not yield a diagnosis. The increased sensitivity of the PCR assay may help to identify those patients with low-titer infections who might benefit from directed antibiotic therapy for P. carinii and would otherwise be missed by direct examination alone.