This is the first meta-analysis that investigates the differential prediction of undergraduate and graduate college admission tests for women and men. Findings on 130 independent samples representing 493,048 students are summarized. The underprediction of women's academic performance (d = 0.14) and the overprediction of men's academic performance (d = -0.16) are generalizable, albeit small. Transferred onto a 4-point grading scale, women earn college grades that are 0.24 points higher than those of men with the same admission test result. Combining admission tests with indicators of previous academic achievements, such as high school grades, reduces the amount of under- and overprediction. Moderator analysis reveals that the underprediction of women's academic performance by admission tests is a problem of the past and present. Predictor differences as well as criterion differences are not associated with over- and underprediction. Rather, undergraduate college admission tests show more underprediction of women's academic performance than graduate admission tests. These results point to differences between undergraduate and graduate students, the latter being more selected. (Contains 3 tables and 3 footnotes.)