Various operational definitions of test bias are discussed including those which involve rater judgment, statistical methods which use an external criterion, and statistical methods in which no external criterion is used. It is concluded that the Rasch model provides a theoretically acceptable method for detecting item bias. The Rasch Model is used to analyze tenth-grade reading test results on the Michigan Educational Assessment Program. Test bias was suspected because White students mastered 14 of the 15 objectives while Black students mastered only 10. Rasch analysis showed no statistically significant bias for the entire test, although one objective (locating items in a category) was more difficult for Blacks than for Whites at the same achievement level, and another (using graphs) was more difficult for Whites. When both objectives were removed the effects were cancelled. It is concluded that it may be more productive to accept test results and concentrate instead on eliminating the inequities which influence the achievement disparity. (Author/CP)