We compared the sensitivities and specificities of four nested PCR assays for the detection of Mycobacterium tuberculosis from formalin-fixed, paraffin-embedded tissues. Thirty-seven autopsy samples from human immunodeficiency virus-positive patients were analyzed: 15 were M. tuberculosis positive, 11 served as negative controls, and 11 were Ziehl-Neelsen positive without cultural confirmation of M. tuberculosis. Three genomic sequences (mtp40, 65-kDa antigen gene, and IS6110) with different molecular masses and numbers of repetitions within the M. tuberculosis genome were targeted. On the IS6110 sequence, two fragments of different sizes (106 and 123 bp, respectively) were amplified with two separate pairs of primers. The highest sensitivity rates were obtained by amplifying the highly repetitive IS6110 insertion sequence, and the different primers tested showed a sensitivity ranging from 80 to 87%. Amplification of the large 223-bp fragment of themtp40 sequence present in a single copy in the M. tuberculosis genome yielded a high rate of false-negative results, ranging from 66 to 80%. A poor sensitivity (from 47 to 60%) was also shown by PCR amplification of the 142-bp 65-kDa antigen gene. All the PCRs except that for the 65-kDa antigen gene showed a specificity of 100%. Moreover, different results were obtained with different dilutions of DNA, and DNA concentrations of 1 and 3 μg yielded the highest sensitivities depending upon which protocol was used. Application of the PCRs to the Ziehl-Neelsen-positive, culture-negative samples confirmed the sensitivities of the PCRs obtained with the control samples. In conclusion, PCR can successfully be used to detect M. tuberculosis from paraffin-embedded tissues and can be particularly useful in the validation of a diagnosis of tuberculosis in clinical settings in which the diagnosis is uncertain. However, the efficacy of PCR strictly depends on several amplification parameters such as DNA concentration, target DNA size, and the repetitiveness of the amplified sequence.