We study workplace segregation in the United States using a unique matched employer-employee dataset that we have created. We first present measures of workplace segregation by race and ethnicity, using simulation methods to measure segregation beyond what would occur randomly as workers are distributed across establishments. We then assess the role of skill differentials in generating workplace segregation, as skilled workers may be more complementary with other skilled workers than with unskilled workers, and skill is often correlated with race and ethnicity. Specifically, we measure the extent of segregation by education level and quality as well as by language ability, providing informative contrasts with measures of segregation by race and ethnicity. Finally, we attempt to distinguish between segregation by skill based on general crowding of the unskilled into a narrow set of jobs versus segregation based on common language for reasons such as complementarity among workers speaking the same language. Our results consistently indicate that workplace segregation by race and ethnicity is driven in large part by skill differences across workers.