The polymerase chain reaction (PCR) in cerebrospinal fluid was compared with conventional bacteriology and an enzyme-linked immunosorbent assay (ELISA) for cerebrospinal fluid antibodies in the diagnosis of tuberculous meningitis (TBM). PCR was the most sensitive technique; it detected 15 (75%) of 20 cases of highly probable TBM (based on clinical features), 4 (57%) of 7 probable cases, and 3 (43%) of 7 possible cases. ELISA detected 11 (55%) of the highly probable cases and 2 each of the probable and possible cases. Culture was positive in only 4 of the highly probable cases. Among the controls (14 pyogenic meningitis, 3 aseptic meningitis, 34 other neurological disorders), 6 subjects tested early in the study (2 pyogenic meningitis, 4 other disorders) were PCR positive. Second DNA preparations from their stored cerebrospinal fluid samples were all PCR negative, suggesting that the false-positive results were due to cross-contamination. 18 PCR-positive TBM samples retested were all still PCR positive. The antibody ELISA was positive in 3 controls despite the use of a high cutoff value.