Exploiting the randomized expansion of preferential college admissions in Chile, we show they increased admission and enrollment of disadvantaged students by 32%. But the intended beneficiaries were nearly three times as many, and of higher average ability, than those induced to be admitted. The evidence points to students making pre-college choices that caused this divergence. Using linked survey-administrative data, we present evidence consistent with students being averse to preferential enrollment, misperceiving their abilities, and having social preferences towards their friends (although social preferences did not mediate the admission impacts). Simulations from an estimated structural model suggest that aversion to the preferential channel more than halved the enrollment impacts, by inducing some to forgo preferential admission eligibility, and that students' misperceptions worsened the ability-composition of college entrants, by distorting pre-college investments into admission qualifications. The results demonstrate the importance of understanding high school students' preferences and beliefs when designing preferential admissions.