In the Early Childhood Longitudinal Study Kindergarten Cohort, I find that the gender gap is not uniform across the distribution of math skills and that these quantile-specific gaps vary with age. Specifically, girls at the top of the distribution initially fall behind boys but manage to catch up later. At the same time, girls in the lower parts of the distribution lose ground. In fifth grade, a gender gap of 0.2 standard deviation, about 2.5 months of schooling, is observed across the entire distribution. Overall, these patterns indicate the possibility that low performing girls become worse and vice versa. These results demonstrate important dynamics of the gap that are relevant for policy, but that the mean gap fails to show.