This paper asks whether reducing socioeconomic gaps in college admission exam scores also reduces earnings inequality. A simple framework shows that the answer to this question depends on the exam’s predictive power for students’ earnings returns to college quality. I exploit a redesign of the Colombian national college admission exam to estimate these returns. Low-income students who took the new exam received higher scores and attended better colleges, but they had lower earnings after college. Thus exams with lower score gaps may not reduce earnings inequality if they are poor predictors of which students would succeed at better colleges.