This paper describes the development and investigation of a new approach to determining the existence of bias in multiple-choice test scores. Previous work in this area has concentrated almost exclusively on bias attributable to specific test items or to differences in test score distributions across racial or ethnic groups. In contrast, the present work concentrates ultimately on the behavior of the individual examinee in choosing among the responses to each item. The major approach was to investigate variation in person-fit, that is, the extent to which an individual examinee selects the choices which are more probable based on the examinee's score and the overall popularity of the choices in the population. Four different person-fit measures were computed for random samples representing subpopulations defined by race, sex, score range and geography. There was strong evidence of variation in person-fit between black and white and between male and female examinees, especially in the lower score ranges. However, subsidiary analyses suggested that this outcome did not reflect bias in the scores from the tests analyzed. Nevertheless, the person-fit approach was shown sensitive to group differences and could reveal the presence of bias if it exists. (Author/CTM)