This paper studies the interplay between major returns, preferences, and market design in explaining the gender earnings gap. Information on preferences allows us to estimate the major impacts while accounting for selection on rich sources of preference heterogeneity. We use this to estimate pecuniary returns to majors and fertility impacts. We find substantial heterogeneity in pecuniary and fertility impacts, with sizable gender differences in each. We find that pecuniary returns are most predictive of male gender choices and the opposite is true for females, providing empirical evidence about an important factor that causes a rift in the educational decisions of men and women. While there are sizable within-major gender differences in returns, we find that most of the college graduate gender gap is explained by differences in representation. Against this backdrop, we take preferences as given and consider policy-relevant counterfactuals leveraging the algorithmic rules embedded in a centralized assignment system. Our results suggest that slight gender-specific quotas can substantially reduce the gender earning gap among college graduates with minimal impact on efficiency.