Fundamental questions about the size and characteristics of the homeless population are unresolved because it is unclear whether existing data are sufficiently complete and reliable. We examine these questions and the coverage of new microdata sources that are designed to be nationally representative. We compare two restricted data sources largely unused to study homelessness, the 2010 Census and American Community Survey (ACS), to restricted Homeless Management Information System (HMIS) data, HUD’s public-use point-in-time (PIT) estimates, and the Housing Inventory Count (HIC) at the national and individual level. We also develop a new approach to estimating the size of the sheltered homeless population using linked Census and HMIS microdata. Our analyses suggest that on a given night there are about 400,000 people experiencing homelessness in shelters in the U.S. and about 200,000 people sleeping on the streets, with this latter estimate subject to greater uncertainty. More than 90 percent of those in shelters appear to be counted in the Census, although many are classified as housed or in other group quarters, due largely to ambiguity in the definition of a homeless shelter. This paper lays the foundation for pathbreaking future work with these data on the U.S. homeless population.