Abstract
Equal opportunity requires that persons be selected for desirable positions on the basis of their qualifications. To assess an applicant's qualifications, we must both predict how well he would perform if chosen, and compare his projected performance with that of his rivals. Since we lack direct access to future performance, all such predictions must be based on some past– or present-tense information about the applicants, together with some relevant supporting information. But is any and every way of predicting performance acceptable? Or are some methods of predicting improper even if they are more accurate than any available alternatives? And, if some methods of predicting are improper, which ones are these, and why?