To evaluate the diagnostic performance of a commercially available Mycobacterium tuberculosis PCR assay (Amplicor MTB-ROCHE), 2296 respiratory and nonrespiratory specimens from 2296 patients with suspected tuberculosis (TB) were collected prospectively in an 8-year period. Clinical data for each patient were abstracted, and all samples were examined blindly by direct microscopy, culture, and PCR. M. tuberculosis DNA was detected in 93 of 113 culture-positive samples and in 29 of 38 samples from patients with probable TB. The lowest sensitivity was observed in pleural fluid and abscess aspirates. The sensitivity, specificity, and positive predictive value of the assay were 97.2%, 100%, and 100% for smear-positive specimens and 75.3%, 97.0%, and 47.5% for smear-negative specimens, respectively. The PCR cost per additional correct clinical decision was Euro 2826 but would have declined to Euro 308 if the test was applied only to smear-positive specimens. The overall performance of Amplicor MTB test was excellent for smear-positive, but suboptimal for smear-negative specimens.