A cooperative study was conducted among six laboratories to compare the performance of the Cobas Amplicor (CA) polymerase chain reaction (PCR) system (Roche Molecular Systems, USA) for the detection of Mycobacterium tuberculosis with that of microscopy and culture in routine clinical laboratory diagnosis. A total of 5,221 decontaminated respiratory specimens were tested. The use of an internal control allowed detection of PCR inhibition in 144 (2.8%) specimens. Only two culture-positive samples were CA PCR inhibitory and therefore could not be detected by PCR testing. Of the 333 culture-positive specimens, 278 (83.5%) were positive by the CA PCR. Of the 4,744 culture-negative specimens, 52 (1.1%) were positive by the CA PCR. After analysis of discrepancies, 40 of the 52 culture-negative, CA PCR-positive specimens were classified as true positive. Thus, the overall sensitivities of culture, CA PCR and microscopy were 89.3%, 85.2% and 55.5%, respectively. The overall specificity of the CA PCR was 99.7%. Five of the six centers found similar performances for the CA PCR, with sensitivities ranging from 85.7 to 90.9%. The CA PCR was more sensitive for smear-positive samples, exhibiting overall sensitivities of 96.1% and 71.7% for smear-positive and smear-negative specimens, respectively. These results indicate that the Cobas Amplicor system enables microbiology laboratories with reasonable previous experience in molecular biology testing to perform PCR and to detect Mycobacterium tuberculosis in more than 70% of specimens obtained from infected patients.