Many studies of tuberculosis have defined clusters of patients on the basis of shared DNA fingerprint patterns of their Mycobacterium tuberculosis isolates. Clustering has been equated with recent transmission, and factors associated with clustering have been sought as a guide to population subgroups with high rates of ongoing transmission of M. tuberculosis. Considerable caution should be exercised in conducting and interpreting these studies. Groups of strains may be identical for reasons other than recent transmission, depending, for example, on the stability of the marker and the number of strains in the population over time. Cases actually due to recent transmission may not be seen as clustered if they are new immigrants to the population or if not all cases in the population are included in the study. The amount of clustering seen will depend on the duration of the study. Studies should give precise information on the study setting, the proportion of cases included, the recruitment period and the definition of clustering used. The data on clustering should be disaggregated at least by age, sex and immigration status. To be maximally informative, studies should involve a high proportion of all cases in a population, be conducted in conjunction with conventional epidemiological investigations of contacts (if possible), and should provide information on tuberculosis incidence in the population and on patients' age, sex, human immunodeficiency virus status, drug resistance and social and ethnic group.