In spite of the large literature on labor market discrimination, the quantity of solid evidence on discrimination is relatively limited. This is because evidence of discrimination is difficult to obtain. Two individuals may be treated equally, but this does not prove discrimination unless we can show that the differences in treatment were not justified by differences in productivity. The method most commonly used to identify wage discrimination, the Oaxaca decomposition, is flawed because any omitted variables that are correlated with gender will contribute to the unexplained portion of the wage gap, leading to an over- or under-estimation of wage discrimination. Audit studies provide more direct evidence of differential treatment, but are costly to carry out. Only a small number of studies attempt to measure worker productivity to see if wage differences are justified. This may be because the data needed to measure productivity are difficult to obtain. This paper tests for wage discrimination by gender and race by estimating relative productivity from 2002 Census of Manufacturing data linked to demographic information on workers from Longitudinal Employer-Household Dynamics (LEHD) files. Comparing the estimated productivity ratios to the observed wage ratios, I conclude that females and blacks face wage discrimination in US manufacturing.