The new Roche Cobas Amplicor Mycobacterium tuberculosisassay, which is a semiautomated version of the manually performed Roche Amplicor M. tuberculosis test, was compared to culture and an IS6110-based in-house PCR protocol. A total of 1,681 specimens from 833 patients, including specimen types other than sputum, were tested in parallel by both the in-house PCR and the Cobas Amplicor M. tuberculosis assay. After we resolved discrepant PCR results, the sensitivity, specificity, and positive and negative predictive values for the Cobas Amplicor M. tuberculosis assay were 66.33, 99.71, 94.36, and 97.66%, respectively. The corresponding values for the in-house PCR were 91.08, 99.85, 97.87, and 99.37%, respectively. For culture- and smear-positive specimens, the sensitivity of the Cobas AmplicorM. tuberculosis test was 96.42% (in-house PCR, 100%). If only smear-negative sputum specimens were considered, the Cobas Amplicor M. tuberculosis assay exhibited a sensitivity of 45.45% (in-house PCR, 63.63%) relative to that of culture. With a modified protocol for DNA extraction (washing of samples plus ultrasonication), both PCR methods performed better with gastric aspirates than with sputum samples (sensitivity of the Cobas AmplicorM. tuberculosis assay with smear-negative gastric aspirates, 70.00%; sensitivity of in-house PCR, 90.00%). With dithiothreitol being used for liquefaction of specimens in this study, the Cobas Amplicor M. tuberculosis assay exhibited an inhibition rate of 9.16%. In our view, the new Cobas Amplicor M. tuberculosis test (i) is well suited for typing of smear-positive specimens, (ii) may also be applied to gastric aspirates and other types of specimens if DNA extraction methods are modified appropriately, and (iii) exhibits a sensitivity with smear-negative sputum specimens which makes it recommendable that a minimum of three samples from the same patient be tested.