BACKGROUND: DNA fingerprint patterns of Mycobacterium tuberculosis strains vary within individuals and between epidemiologically linked individuals because of pattern evolution, new infections, and laboratory error. We explored the importance of these factors. METHODS: Cultures from individuals in northern Malawi who had been diagnosed with tuberculosis (TB) during 1996-2001 were fingerprinted with restriction fragment-length polymorphism (RFLP). Probable laboratory error was inferred by use of dates or isolated positive cultures. Pattern evolution was explored within and between individuals, and the relative importance of relapse and reinfection was estimated in individuals with recurrent TB. RESULTS: RFLP results were available for 930 isolates from 806 individuals. The maximum laboratory-error rate was 3.9%. Pattern evolution was more common in linked individuals (17%) than on relapse (11%) or during treatment (3%). Twenty individuals had recurrent TB after completing treatment: in human immunodeficiency virus (HIV)-positive individuals, 7 of 12 recurrences were due to reinfection, compared with 0 of 8 in HIV-negative individuals (P=.01). CONCLUSIONS: The rate of DNA fingerprint-pattern evolution is not linear, and rates of change calculated from repeat cultures within individuals may not be applicable to transmission between individuals. The high proportion of recurrence due to reinfection found in HIV-positive individuals suggests that secondary prophylaxis and/or antiretroviral treatments are needed for such individuals.