A total of 266 Mycobacterium tuberculosis isolates were subjected to DNA RFLP analysis. They were obtained from monthly sputum cultures from patients treated with short-course chemotherapy and then followed up for 2 years. They originated from 42 patients who relapsed after short-course chemotherapy and from a further 42 patients who yielded a single isolated positive culture after chemotherapy. The isolates consisted of one obtained pretreatment and the last obtained during chemotherapy, together with either two isolates cultured at least 2 months apart during relapse or the single post-chemotherapy isolate. They were coded before DNA RFLP analysis and assigned to groups with identical or near identical band patterns on visual inspection. After decoding, it was evident that almost every patient was infected with a strain with a different band pattern (fingerprint). In 100 comparisons of either the pretreatment isolate against the last positive isolate obtained during chemotherapy, or of the first relapse isolate against the second relapse isolate, 15 had been recorded as different; 4 of these were retrospectively found to be due to reading error (error rate 1.5%), leaving 11 (11%) with marked differences. For 5 (12%) of the 42 patients who relapsed, the fingerprint of the relapse isolate was markedly different from that of the pretreatment isolate. In contrast, the isolated positive culture was markedly different from that initially present in 36 (90%) of 40 comparisons. The relative contributions by clinical mixed infection and laboratory cross-contamination to the remaining 10-12% discrepancy rates could not be assessed.(ABSTRACT TRUNCATED AT 250 WORDS)