Various polymerase chain reaction (PCR) assays have been devised for the rapid identification of mycobacteria in clinical specimens. To assess the value of such assays in routine laboratory work the results obtained by PCR were compared with those obtained by standard microbiological methods for 514 specimens collected for investigation of mycobacterial infection. Specimens were tested for the presence of Mycobacterium tuberculosis complex and atypical mycobacteria in two assays, one based on amplification of the 65 kDa gene and the other on the IS6110 insertion sequence. For the 489 samples that did not contain inhibitors of the amplification reaction PCR findings correlated well with bacteriological and/or clinical data in 476 (97.4%). 6 PCR results turned out to be false negatives, 3 to be false positives and 4 to be mis-identification of strains. Pre-treatment of samples with guanidium thiocyanate reduced the proportion of false-negative results and of samples that contained inhibitors. This study confirms the potential of DNA amplification for early diagnosis of mycobacterial infections.