This paper considers the linkage of empirical estimates of wage discrimination between two groups, introduced by Oaxaca (1973), to a theoretical model of employers' discriminatory behavior. It is shown that, conditional on different assumptions about employers' discriminatory tastes, Oaxaca's estimators of wage discrimination can be derived. That the approach is more generally useful is demonstrated by deriving an alternative estimator of wage discrimination, based on the assumption that within each type of labor (e.g., unskilled, skilled) the utility function capturing employers' discriminatory tastes is homogeneous of degree zero with respect to labor inputs from each of the two groups. The estimators are compared empirically in an application to male-female wage differentials. (This abstract was borrowed from another version of this item.)