Accurate assessment of attitudes about intimate partner violence is important for evaluation of prevention and early intervention programs. Assessment of attitudes about cross-gender interactions is particularly susceptible to bias because it requires specifying the gender of the perpetrator and the victim. As it is likely that respondents will tend to identify with the same-gender actor, items and scales assessing attitudes about intimate partner violence may not have equivalent measurement properties for male and female respondents. This article examines data from 2,575 high school students who participated in a teen-dating violence intervention study. The majority of participants were Latino (91%), and the sample was nearly evenly split with respect to gender (51% female). Items from two scales (boy-on-girl violence, 4 items; girl-on-boy violence, 5 items) reflecting teens' attitudes about dating violence were calibrated with the graded item response theory (IRT) model and evaluated for differential item functioning (DIF) by gender. A total of three items, two from the girl-on-boy violence scale and one from the boy-on-girl violence scale, were identified as functioning differently for girls and boys. In all cases where DIF was detected, the item's attitudinal statement was easier to accept for the gender group that was portrayed as victim rather than perpetrator. For both scales, accounting for the identified DIF influenced inferences about the magnitude of mean differences in attitudes between boys and girls. These results support the use of IRT scores that account for DIF to minimize measurement error and improve inferences about gender differences in attitudes about dating violence. (Contains 2 figures, 4 tables and 2 notes.)