The purpose of this study was to evaluate the reliability of an amplification restriction analysis based method (hsp65-RFLP) to detect and identify mycobacterial species in clinical samples and cultures with low number of bacilli. We examined 247 clinical specimens and 88 culture vials, comparing hsp65-RFLP results with conventional culture/biochemical tests. The analytical sensitivity of the method was assessed with cerebrospinal fluid (CSF), broncho-alveolar lavage (BAL), sputum, water, and 12B medium containing defined amounts of mycobacterial chromosome. We detected the equivalent of 10(3) cells per ml in all samples, except sputum, the most common source of clinical sample for mycobacterial testing, which presented inhibition throughout. We investigated two purification procedures to overcome inhibition of DNA amplification: DNAzol and phenol/chloroform. The former was superior, eliminating inhibition in 93.7% of the clinical samples. The technique was effective for bacterial cultures, including those with very low growth indices (GIs), substantially abbreviating time for diagnosis, but showed low sensitivity (25%) when applied to clinical samples, an issue that has never been extensively assessed by other researchers.