The effectiveness of the Waksberg telephone sampling method and practical aspects of its implementation were evaluated during the selection of control subjects for a population-based case-control study conducted in Maryland, Virginia, West Virginia, and Washington, DC, in 1991. The first stage of the two-stage sampling procedure was equivalent to simple random sampling. Each telephone number selected was dialed up to 13 times to establish the type of connection. Residential connections were used to generate additional telephone numbers in the second stage by changing the last two digits. Overall, 3,860 telephone numbers yielded 1,311 residences. The proportion of residences was higher in the second stage (54%) than in the first (11%), while that of numbers "not in service" was lower (12% vs. 67%, respectively). Fewer calls (8,735) were made than would have been required by a simple random sampling procedure (up to 18,522). The geographic distribution of residences was similar at both stages of sampling. At the end of enrollment, each sampling unit had yielded 9.8 residences, but only 64% reached 9-11 residences. Compliance rates were 96% for screening (1,259 of 1,311 residences), 90% for interviewing (362 of 403), and 86% overall. Relative to simple random sampling, the Waksberg method considerably reduced the total number of telephone calls, but the sampling process was more complex to manage. This method is particularly suitable for selecting population controls when the density of residential connections is low and variables of interest are not homogeneous within sampling units. Random digit dialing remains an effective method for selecting population controls.