The U.S. Census Bureau conducted the Ethnographic Evaluation of the Behavioral Causes of Census Undercount for the 1990 Decennial Census. The evaluation consisted of intensive studies of twenty- nine sample areas conducted by ethnographers. This paper explores a way to define the effect of the twenty-eight sample areas located in the continental U.S. The approach taken is to group the sample areas into clusters, using the site-level data from the Census long questionnaires on various aspects of the sample areas, including their social, economic, and educational backgrounds. The newly defined clusters will be included in the later analyses as one of the possible factors that best explain the census coverage errors.