As part of a program of coverage research in th 1990 census, the Census Bureau sponsored ethnographic studies to investigate hypothesized behavioral causes of coverage errors, including language and illiteracy; mobility; irregular housing and household arrangements; resistance as a strategy for dealing with outsiders, especially Government; and concealment to protect resources combined with disbelief in confidentiality (Brownrigg and Martin, 1989). In 29 sites in which undercounts were expected to be high, the Census Bureau contracted with ethnographers to conduct their own independent enumerations of the sites, which were then compared with census results, with followup fieldwork to resolve and explain discrepancies. Their research provides rich information about factors which contributed to omissions and erroneous enumerations in the census (de la Puente, 1993). Drawing on the ethnographic research, this paper describes some types of households and living arrangements which contributed to coverage errors in the census. The census attempts to enumerate persons at their "usual resident," where they live and sleep most of the time. However, some people do not have a usual residence so defined, and for others usual residence may be ambiguous or uncertain. Still others may not understand or follow the instructions on the census form about who should be included on the roster and who left off. The Census Bureau is examining the concept of usual residence and how it is implemented in the census in a research program which includes collection of basic data on residency patterns and household attachments, as well as experimentation to develop and test alternative roster methods to improve coverage. Part of this program of research is the Living Situation Survey, which is being conducted by RTI May-August 1993 under contract with the Census Bureau. One thousand households, and 2,200 individuals, will be interviewed in a sample which disproportionately represents areas with high concentrations of minorities and renters, who are known to be undercounted in the census. The Living Situation Survey employs new, experimental methods to list a much more inclusive household roster by using probes designed to learn about persons with a variety of attachments to the household. The LSS collects complete information about actual residency patterns from individuals, by using calendars to record where people stayed for the previous 2-3 months. These objective data will be used to determine how well census residency rules apply to actual living situations, and to determine (within sample limitations) the proportion of rostered individuals who had no usual residence, or for whom usual residence is ambiguous. Analysis of the LSS data will permit us to identify characteristics of persons marginally attached to households, and will provide information about how roster methods might be changed to improve coverage of this group. We will assess how well household respondents' reports of usual residence agree with the objective facts of individuals' own reports of their living situation. We will assess the evidence to assess whether de facto enumeration, that is, counting people where they are found rather than at their usual residence, might lead to improved coverage of the population. Finally, the survey will allow us to identify ambiguities and problems in terminology and the application of the rules, and to improve roster instructions.