Typically, group differences are analyzed at the subdomain or test level using composite scores. This can mask the effect of individual items across groups. For example, two items from the Physical Self-Description Questionnaire (PSDQ) are worded in terms of internal ("I am good looking") and external ("Nobody thinks Im good looking") frames of reference. Any difference between the group means using a composite score including these two items is interpreted globally, thus obscuring individual item meaning. A better approach to detecting group differences is to work at the item level using differential item functioning (DIF) methods. One such method is multidimensional DIF, which identifies any secondary unintended latent dimensions at the item level that may lead to between-group score differences. The results suggest that 28 of the PSDQ's 70 items (40%) are biased in favor of either men (17%) or women (23%) and that a primary source of DIF relates to item composition (negative wording) and internal or external frames of reference.