This study was designed to increase knowledge of the functioning of item bias techniques in detecting biased items. Previous studies have used computer-generated data or real data with unknown amounts of bias. The present project extends previous studies by using items that are logically generated and subjectively evaluated a priori to be biased or unbiased, and simultaneously controls the amount of bias and true ability differences (as measured by the unbiased items). The study evaluated the functioning of four statistical methods of assessing test item bias (transformed item difficulties, chi-square, three parameter and one parameter item characteristic curves) when (1) tests have varying amounts of bias (0 biased/60 items, 18 biased/78 items, 40 biased/100 items) and (2) ability differences on the unbiased items were either one-half or one standard deviation apart. Results indicate that agreement among the methods and between the statistical methods and judged bias was generally high except for data set VI (40 percent biased items, one standard deviation difference). Problems with individual methods and with cutoffs are discussed. Finally, presence of biased items did not affect reliability but did decrease validity and increase score differences between groups. (Author/CM)