A number of methods exist for detection of differential item functioning (DIF), but these methods tell us little about the causes of DIF. DIF is typically defined based on a relationship with some manifest group characteristic, such as gender or ethnicity, which is only weakly associated with DIF. What is lacking is a method that will lead to understanding why some examinees respond differently to an item and others do not. In this study, item response patterns on a set of 1996 National Assessment of Educational Progress grade 8 State Mathematics Assessment items for ethnic group DIF were analyzed in two different samples. In one sample, responses from Caucasian and African American examinees were compared, and in a second sample, responses from Caucasian and Hispanic examinees were compared. The patterns of DIF in these two samples were used to define two separate mixture item response theory models in an attempt to determine whether the manifest characteristics of ethnic group membership was useful in identifying those examinees who were actually advantaged or disadvantaged by the item(s) in either sample. Comparisons of background and attitude variables of these examinees revealed some clear differences between manifest and latent groups. (Contains 4 figures, 7 tables, and 21 references.) (Author/SLD)