An item content criterion, independent of the test psychometric characteristics, of classifying items as biased is described. It was used with a sample of female adults in a training program for administrative secretaries. The minority group in the study were Hispanic immigrants. The majority group was a mixture of Blacks, English speaking Caribbeans, and Hispanics born in the continental United States or those who attended American schools. The impact of removing biased items on the validity coefficient and the groups' mean scores was investigated. The elimination of biased items reduced the mean difference in the predictor between the majority and minority groups, thus improving the test's validity. The most discrepant items by the Transformed Item Difficulty (TID) method and the items with unacceptable difficulty differences between the groups were compared with the items detected as biased by the item content criterion. The greatest similarity was found between the item content criterion and the unacceptable difficulty differences. (Author/DWH)