AIMS: To evaluate the usefulness of a single-tube nested polymerase chain reaction (PCR) assay in the diagnosis of tuberculosis in 1497 pulmonary and 536 extrapulmonary specimens. METHODS: A single-tube nested PCR, utilising two sets of primers with different melting temperatures (88 degrees C for external primers; 70 degrees C for internal primers) to augment sensitivity and specificity without increasing the risk of amplicon contamination, was evaluated. Specimens were initially tested for the repetitive IS6110 sequences and if negative, retested for the universal 38 kilodalton sequence and for inhibitors. dUTP/Uracil-N-glycosylase and Instagene treatment were used to minimise contamination and the effect of inhibitors, respectively. RESULTS: Using culture as the gold standard, the overall sensitivity of the assay was 89% for pulmonary and 42% for extrapulmonary specimens. Sensitivity varied greatly with respect to sample type (92% for follow up specimens from a chest hospital and 70% for non-follow up specimens from a general hospital). The smear positivity rates were 15% for extrapulmonary specimens, and 69% and 45%, respectively, for follow up and non-follow up specimens from pulmonary sites. Specificity was 99.7%. Inhibitors were present more frequently in extrapulmonary than in pulmonary specimens (13.4% v 2.7%). CONCLUSION: Despite the high sensitivity of the PCR assay for the diagnosis of tuberculosis in pulmonary specimens, it was less effective in the extrapulmonary samples. This is probably because of the lower bacterial load in extrapulmonary specimens, the presence of more inhibitors adversely affecting the PCR assay and the higher volume of specimens used for culture.