Using a unique dataset, this article first documents that gaps in starting wages by race and sex persist after accounting for performance on the job. Evidence suggests that simple statistical discrimination, and not just taste discrimination, is partly responsible for race differences in starting wages. But because women's average performance in the sample is higher than men's, simple statistical discrimination cannot explain the sex gap. In more complex models of statistical discrimination, worse information about a group can lower its average wage. Estimates of the quality of labor market information indicate that this may explain women's lower starting wages.