The reliability of the Roche AMPLICOR Mycobacterium tuberculosis test (AMPLICOR MTB) for the diagnosis of pulmonary tuberculosis was evaluated by testing 956 respiratory specimens from 502 patients and comparing results with results by culture and medical history. Of those 135 specimens that were culture positive for mycobacteria, 61 specimens from 31 patients grew M. tuberculosis. Fifty-two specimens were smear positive for acid-fast bacteria (AFB); M. tuberculosis was isolated from 41 of these specimens. On initial testing, the sensitivity and specificity of the AMPLICOR MTB assay, compared with culture, were 78.7 and 99.3%, respectively. After resolution of discrepancies (by review of medical history), the sensitivity, specificity, and positive and negative predictive values of the AMPLICOR MTB assay were 79.4, 99.6, 92.6, and 98.6%, respectively. Two specimens from two patients with no clinical evidence of tuberculosis were AMPLICOR MTB positive and culture positive for Mycobacterium avium complex. For AFB smear-positive specimens, the sensitivity, specificity, and positive and negative predictive values of AMPLICOR MTB were 97.6, 100, 100, and 90.9%, respectively. For AFB smear-negative specimens, the sensitivity, specificity, and positive and negative predictive values of AMPLICOR MTB were 40.0, 99.5, 69.2, and 98.7%, respectively. Our results support the use of AMPLICOR MTB for rapid diagnosis of tuberculosis in patients whose respiratory specimens are AFB smear positive. Further studies are needed to determine the most clinically relevant and cost-effective use of this assay with AFB smear-negative specimens.