A procedure for the detection of differential item performance (DIP) is used to investigate the relationships between characteristics of mathematics achievement items and gender differences in performance. Eight randomly equivalent samples of high school seniors were each given a unique form of the ACT Assessment Mathematics Usage Test (ACTM). Students without requisite math courses were deleted from the samples to control the possible confounding effect of differences in instruction at the high school level. Based on the remaining students, signed measures of DIP were obtained for each item in the eight ACTM forms. These DIP estimates were then investigated in a six by eight (item category by firm) experimental design. Using analysis of variance (ANOVA) procedures, a significant item category effect was found indicating a relationship between item characteristics and gender-based DIP. Follow-up analyses suggested that geometry and mathematics reasoning items had the largest negative impact on female examinees and more algorithmic, computation-oriented items were relatively easier for females. (Author)