We evaluated the Amplicor PCR assay (Roche Molecular Systems, Branchburg, N.J.) for direct detection of Mycobacterium tuberculosis in sputum. A total of 532 specimens from 270 patients were decontaminated and stored at 4 or ؊75؇C until assayed by PCR. This assay used three-step sample preparation, biotinylated primer pairs, AmpErase, and a microtiter format for amplicon capture and detection. Amplicor PCR results were compared with clinical history, culture from a Lowenstein-Jensen slant, and results from the BACTEC TB-460 system. Eighty-seven cultures from 15 patients grew M. tuberculosis; of these, 83 (95%) were positive with the Amplicor PCR test. The false negatives were most likely due to sample variation and inhibitors. Of the 445 specimens from which M. tuberculosis was not isolated, 428 (96%) were negative with the Amplicor PCR test. Of the 17 M. tuberculosis culture-negative, Amplicor-positive specimens, 15 were reclassified as true positives because previous cultures grew M. tuberculosis. Of the 445 specimens which did not grow M. tuberculosis, Mycobacterium spp. other than M. tuberculosis were isolated from 150 specimens. Three of these 150 specimens were Amplicor positive; two were from a patient with a history of tuberculosis, and one specimen gave a false-positive result. We do not feel that this represents cross-reactivity, because repeated Amplicor testing of the isolate gave negative results. The microtiter plate has 96 wells. Allowing for six controls, 90 decontaminated specimens can be tested by one technologist in 7.5 h. This PCR assay took 7.5 h to complete and is a sensitive and specific, rapid method for the direct detection of M. tuberculosis from sputum.