This study examines the diagnostic utility of the polymerase chain reaction (PCR) in 156 patients (five human immunodeficiency virus (HIV) seropositive) suspected of extrapulmonary tuberculosis. The results of PCR in 226 samples from 11 different sites were compared with the results of microscopy and culture. Positive culture results were predicted in 86% of samples by PCR but in only 31% by microscopy. Specificity of PCR was 92%. In cases with culture-proven tuberculosis, PCR identified all 11 microscopy positive cases and 19 of 24 (79%) of the microscopy-negative cases. In four patients, PCR excluded the diagnosis of tuberculosis in microscopy-positive samples, which were later shown to contain mycobacteria other than Mycobacterium tuberculosis or laboratory contaminants. In 20 patients (microscopy, PCR and culture negative) a trial of antituberculous drugs was given, but patients showed no improvement and treatment was stopped. In 17 patients, all culture negative (in nine PCR was positive, three of whom also had positive microscopy) the diagnosis was probable tuberculosis based on clinical findings and response to treatment. This polymerase chain reaction has a much higher sensitivity than microscopy and can facilitate therapeutic decisions for those with suspected extrapulmonary tuberculosis.