This dissertation consists of two chapters studying the importance of household income for shaping student outcomes in the market for higher education in the United States. The first chapter uses the High School Longitudinal Study of 2009 to document that conditional on student ability, high-income students are more likely to enroll in college and are more likely to attend a highly selective college conditional on enrolling. These gaps are mostly explained by differences in application rates and in enrollment rates conditional on being admitted, rather than differences in admission rates. While students generally prefer to attend the most selective college they are accepted to, low-income students are less likely to attend their preferred college due to costs. These findings suggest that financial aid provided by colleges is generally insufficient in closing enrollment gaps, and that the observed application gaps may be rational: low-income students will choose not to apply if doing so is costly and they do not expect to receive sufficient aid if admitted. Motivated by the empirical findings of the first chapter, the second chapter builds and estimates an equilibrium model of the U.S. college market featuring tuition discrimination and a decentralized admissions system. Students who differ in their financial resources and innate ability apply to a subset of colleges and are uncertain about their prospective admissions and financial aid. Colleges observe a noisy signal of student ability and compete by choosing admissions standards and tuition schedules. According to the estimated model, differences in application rates between high- and low-income students, conditional on ability, are due to student expectations over admissions and financial aid, which are consistent with college policies in equilibrium. Low-income students receive generous financial aid at selective colleges because only the highest-ability among them apply, making their signals highly informative. If signals became less informative (e.g., colleges stopped using the SAT), all high-ability students would be worse off and only high-income, low-ability students would modestly benefit. Finally, the model suggests that increasing federal need-based financial aid greatly benefits low-income, high-ability students by alleviating credit constraints.