Choosing an appropriate sampling frame for an epidemiologic study depends on the characteristics of the population under investigation and the sources available for identification of potential subjects. Not all countries maintain (or permit sampling from) population registers. In studies of dispersed populations, electoral rolls may lead to sampling bias, and random-digit dialing is inefficient. A dispersed population characterized by distinctive surnames might be appropriately identified and sampled through telephone directory listings, however. The method proved feasible and efficient for a prevalence study of health-related characteristics of Chinese residents in Melbourne, Australia, yielding a sample demographically comparable to the corresponding census population.