In previous studies of ethnic group bias in intelligence test items, the question of bias has been confounded with ability differences between the ethnic group samples compared. The present study is based on a conditional probability model in which an unbiased item is defined as one where the probability of a correct response to an item is the same for all persons of a given ability level regardless of that person's ethnic group background. Using this definition and the chi square procedure for isolating biased items developed by Scheuneman, a large pool of group-administered intelligence test items were examined for ethnic group bias using samples of whites, blacks, and Mexican-Americans. When assessed at grades 8-9, 68 out of 480 items were found to be biased. At grades 11-12, 79 out of 240 items appeared to be biased. In examining the direction of bias, clear evidence of bias against blacks, but no evidence of bias against Mexican-Americans was found. At both levels, the highest proportion of biased items was among the verbal analogy items and letter and figural series completion items. (Author/RD)