Detection of Mycobacterium tuberculosis in clinical specimens by the polymerase chain reaction (PCR) was compared with detection by culture. A 317-bp segment within the M. tuberculosis-specific insertion sequence IS6110 was amplified. The detection limit of the PCR assay for cultured mycobacteria was 50 cells per reaction by ethidium bromide-stained agarose gel electrophoresis and 5 cells per reaction by hybridization with an oligonucleotide probe conjugated with either digoxigenin or alkaline phosphatase (AP). This sensitivity was reduced fivefold in sputum specimens seeded with M. tuberculosis. Seventy-six clinical specimens were amplified and examined by the three detection methods. Both the digoxigenin and AP procedures were found to be more sensitive than agarose gel electrophoresis, but they were occasionally associated with a high background. An additional 308 specimens were examined only by agarose gel electrophoresis and the AP procedure. Of 71 specimens found to contain M. tuberculosis, amplified products were detected from 56 (79%) samples by agarose gel electrophoresis and/or the AP procedure. Of the additional 313 specimens that were culture negative for M. tuberculosis, 19 (6%) had amplified products detectable by agarose gel electrophoresis and/or the AP procedure. Compared with culture, PCR showed sensitivities and specificities of 55 and 98%, respectively, for agarose gel electrophoresis and 74 and 95%, respectively, for the AP procedure. Despite this low sensitivity, a rapid positive PCR result was accurate and clinically useful.