A test bias definition, applicable at the item-level of a test is presented. The definition conceptually equates test bias with measuring different things in different groups, and operationally equates test bias with a difference in item fit to the Rasch Model, greater than one, between groups. It is suggested that the proposed definition avoids confusing etiology with measurement by capitalizing on the "objectivity" property of the logistic Rasch Measurement Model. Application of the definition, to 914 applicants (black = 367; white = 547) in a "real" selection situation is described and resulted in identifying two items as biased. The two items so defined, were different than the two items identified as biased by comparing the item success rates (i.e., item difficulty) of black vs. white candidates. A content evaluation of the items by two black, female reviewers was subsequently performed. Their comments lend preliminary support to the proposed psychometric test bias definition. Additional encouraging support is provided by the match between the content comments and the item bias index values, for other items in the test. Implications for future applications and research are presented. (Author)