A large differential undercount biases census and survey data collected by the Census Bureau. This paper reviews evidence on magnitude and causes of undercount. A recent ethnographic study and other sources of evidence are used to evaluate two sources of undercount. Undercount results if people deliberately choose not to be counted, or if their living arrangements are incompatible with the Census Bureau concept of "usual residence." Both causes result in biased estimates of important social characteristics. We recommend research in four areas: participant observation studies of causes of undercoverage; research on household structure; comparisons of matched data from the Current Population Survey (CPS) and the census; and research on interviewer training and performance.