This paper presents findings from research exploring gender by item difficulty interaction on mathematics test scores in Cyprus. Data steamed from 2 longitudinal studies with 4 different age groups of primary school students. The hypothesis that boys tended to outperform girls on the hardest items and girls tended to outperform boys on the easiest items was generally supported for each year group. The effect of social class was also examined. For each social class, there was a correlation between the item difficulty differences estimated on girls and boys separately and the difficulty of the item estimated on the whole sample. It is claimed that in understanding gender differences in mathematics, item difficulty should be treated as an independent variable. Suggestions for further studies are provided, and implications for the development of assessment policy in mathematics are drawn. (Contains 2 tables and 8 figures.)