Two competing conceptualizations of test bias in educational settings were studied and each was compared to the traditional regression model. The bias model proposed by Clcary and that proposed by Thorndike were each employed, along with the traditional regression model, on data obtained from 95 male and 101 female fifth-grade children to determine the extent of sex bias in selection resulting from each of the three models. It was found that the Cleary procedure produced substantial bias in both selection and accuracy of selection in favor of the female subgroup. The traditional prediction model resulted in pronounced selection bias in favor of the male subgroup and a slight bias in accuracy of selection in favor of the same subgroup. The Thorndike procedure, by definition, results in no selection bias but did produce slight bias in the accuracy of selection favoring the female subgroup.