In this paper we examine four alternative methods for estimating the e
xtent of labor market discrimination. All of the methods involve the d
ecomposition of gross (unadjusted) wage differentials into discriminat
ion and productivity components. These methods can be expressed in a s
ingle generalized form and are shown to differ with respect to the imp
licitly assumed nondiscriminatory, competitive wage structure. Equival
encies among the methods are shown to exist under certain restrictive
conditions. These methods are applied to micro data from the U.S. Curr
ent Population Survey and from a specific U.S. firm.