We have compared the sensitivity and specificity of quantitative mycobacterial culture against results obtained by using the polymerase chain reaction for the detection of DNA from organisms of the Mycobacterium tuberculosis complex in 82 clinical specimens from patients suspected of having tuberculosis. Two amplification protocols were used, a standard amplification protocol, which amplifies a segment of the gene coding for the 65-kDa antigen, and a protocol in which the initial amplification products are reamplified with a second set of nested oligonucleotide primers. Although the standard amplification protocol gave positive results for 18 of 18 samples which grew greater than 100 CFU/ml and gave positive results in 4 of 35 specimens from patients with tuberculosis which were negative by culture, only 1 of 6 samples which grew less than 100 CFU/ml was positive. This lack of sensitivity could not be explained by the presence of inhibitors of Taq polymerase present in the original samples. In contrast, the reamplification protocol gave positive results for 24 of 24 samples which were positive by culture as well as for 13 of 35 samples from patients with tuberculosis which were negative by culture (overall sensitivity, 63%, P less than 0.02, compared with the standard amplification protocol and routine culture). Two of 23 samples from patients not diagnosed as having tuberculosis gave positive results when the standard amplification protocol was used, but no additional false-positive results were seen with the reamplification protocol (overall specificity, 91%). We conclude that the use of a reamplification protocol improves the sensitivity of detection of mycobacterial DNA in clinical samples without sacrificing specificity. The sensitivity of this approach appears to be superior to that of standard culture techniques.