The use of "bias elimination procedures" to reduce the racial bias of test items is discussed. These procedures were forwarded by G. R. Anrig (1988) and R. L. Linn and F. Drasgow (1987). Anrig stated that subjects who "know the same amount about a test item" should have a similar chance of answering it correctly "regardless of their race, sex, or ethnic background." Linn and Drasgow stated that an adequate approach to detecting item bias "requires a means of distinguishing between differences that are due to group differences in the developed skills of the test takers and those that are due to extraneous factors." The latter researchers propose a one-dimensional item response theory (IRT) criterion. However, this procedure provides no agreed external criterion for making a judgment concerning bias. A requirement to select those items that minimize group differences on the final test does appear to meet a general requirement for equity; this is the procedure forwarded by the Golden Rule Insurance Company in its debate with the Educational Testing Service. It is concluded that: (1) there may be no purely technical solution to the problems of test bias; (2) the test construction process should recognize the need to make ideological and social choices; and (3) IRT theory will not provide a solution. (TJH)