During the fall of 1992, NORC carried out a methodological experiment for the U.S. Bureau of the Census. The purpose of the study was to test new versions of the roster questions used to enumerate members of sample households in many surveys in an effort to improve coverage of household members. In addition, the new roster questions permitted comparison of different rules for linking persons to dwellings. In 1991, a test of differing rostering procedures had been conducted in conjunction with the American Housing Survey (Shapiro, Diffendal, and Cantor 1993). In the same year, a small-scale We carried out an experimental comparison of three versions of questions for enumerating the residen ts of a dwelling. One version took the approach that is used in many surveys: it began by asking respondents to name all persons living at the dwelling. The experimental versions began by asking how many persons had spent the previous night at the dwelling and used other probes to complete the roster. The two experimental versions differed only in that one version did not require persons to be listed by their full names, allowing respondents to use initials or nicknames instead. A total of 509 interviews were completed, about a third of them with each version of the questionnaire. The results indicated that both experimental versions of the roster questions yielded more persons per household than the standard version; however, only the version that did not require full names yielded more persons identified as usual residents of the dwelling. Additional analyses indicated that the same types of persons were listed on all three versions; only the number of persons listed differed.