The roster of individuals living or staying in a household is a fundamental component of most household surveys and censuses. Obtaining accurate and complete rosters from respondents is critical for producing complete population counts and reliable statistics on household composition and family characteristics. Recognizing that rostering errors might occur because of a lack of fit between census rostering instructions on whom to list and the way respondents conceptualize household membership, the Census Bureau sponsored the Living Situation Survey, in part, to explore factors respondents use in determining household membership. The aim of this paper is to use these survey data to compare respondents' subjective judgments about rostered persons' household membership status with objective reports of each persons' mobility over time and social attachments to the household. Intercorrelations among fourteen measures of social attachment to households are examined. A composite measure of household attachments is proposed. Correlations between household and individual respondents' determinations of usual residence are examined. The findings are related to anthropological perspectives on the "household." Implications for coverage research and for application of the method, "collect de facto, tabulate de jure," are discussed. Recommendations for rostering methods and questions to use in the 1995 test census are presented.