The purpose of this investigation was to examine the interaction of item content and group membership on achievement test items. Estimates of the parameters of the three parameter logistic model were obtained on the 46 item math test for the sample of eighth grade students (N = 2055) participating in the Illinois Inventory of Educational Progress, Illinois' statewide assessment booklet. Black students were divided into quintiles on the basis of their estimated theta's. Average standardized difference scores were computed within each quintile as well as across all black students. Some items were identified as biased in favor of black students because their actual performance is systematically better than expected while some were considered biased against black students relative to other items on the math test. Finally, items were classified by content categories and compared in terms of their average standardized difference score. Questions about the metric system, those involving definitions, and those based upon graphs and figures stood out as ones on which black students did worse than expected from the latent trait model. Story problems involving money, those involving unknown symbol substitution stood out in the other direction. Recommendations are made for test revision and for a line of inquiry into what is now labeled as item bias. (Author)