A frequently observed research finding is that females outperform males on tasks of verbal and language abilities, but there is considerable variability in effect sizes from sample to sample. The gold standard for evaluating gender differences in cognitive ability is to recruit a large, demographically representative sample. We examined 3 decades of U.S. student achievement in reading and writing from the National Assessment of Educational Progress to determine the magnitude of gender differences (N = 3.9 million), and whether these were declining over time as claimed by Feingold (1988). Examination of effect sizes found a developmental progression from initially small gender differences in Grade 4 toward larger effects as students progress through schooling. Differences for reading were small-to-medium (d = −.32 by Grade 12), and medium-sized for writing (d = −.55 by Grade 12) and were stable over the historical time. Additionally, there were pronounced imbalances in gender ratios at the lower left and upper right tails of the ability spectrum. These results are interpreted in the context of Hyde’s (2005) gender similarities hypothesis, which holds that most psychological gender differences are only small or trivial in size. Language and verbal abilities represent one exception to the general rule of gender similarities, and we discuss the educational implications of these findings.