UNLABELLED: Reports on the sensitivity of polymerase chain reaction (PCR) for the diagnosis of lymph node tuberculosis (TB) show divergent results. We evaluated the accuracy of the Roche Amplicor Mycobacterium tuberculosis PCR test with lymph node aspirate and biopsy samples. METHODS: The study was conducted at a public reference hospital in Lima, Peru. From the period of January 2003 to January 2004, we included patients who had lymphadenopathy and in whom the attending physician suspected TB. Aspirate and biopsy samples were submitted for culturing in Lowenstein-Jensen medium, for histopathologic testing, and for PCR. The sensitivity and specificity of PCR were calculated against a reference standard based on histopathologic findings and culture. RESULTS: Our study included 154 patients. Median age was 29 years (interquartile range, 21-40 years); 97 patients (62.9%) were men. Twenty-nine patients (18.8%) had acid fast bacilli-positive histopathologic findings, and 44 (28.6%) had a positive culture result. Using the combination of histopathologic findings and culture as reference standard, 55 patients (35.7%) had a diagnosis of tuberculous lymphadenitis. The sensitivity of the PCR test was 58.2%, and the specificity was 93.9%. For biopsy tissue only, the sensitivity of PCR was 52.7%, and the specificity was 97.0%. For aspirate samples only, the sensitivity of PCR was 47.3%, and the specificity was 96.0%. CONCLUSION: The Amplicor PCR test revealed low sensitivity and high specificity for the diagnosis of lymph node TB. The sensitivity was higher in cases in which the bacillary load was high--in acid fast bacilli-positive samples and among HIV-infected patients. Considering the results of microbiological and PCR tests together, there was still a patient group in whom no final diagnosis could be established.