The agreement between item response theory-based and Mantel Haenszel (MH) methods in identifying biased items on tests was studied. Data came from item responses of four spaced samples of 1,000 examinees each--two samples of 1,000 Anglo-American and two samples of 1,000 Native American students taking the New Mexico High School Proficiency Examination in 1982. In addition, a matched group analysis was conducted using a third sample of 650 Native Americans and 650 Anglo Americans. The item characteristic curve area and the MH methods were used. The consistency of classification of items into biased and not-biased was in the 75 to 80% range for both methods. When the unreliability of item bias statistics was taken into account, both methods gave similar results. Discrepancies between methods were due to bias from intersections of item characteristic curves and the choice of interval over which item bias was defined. The Mantel-Haenszel method, with a minor modification or two, provides an acceptable approximation to the item response theory based methods. Five data tables and eight graphs show study results. (Author/SLD)