Improving college performance and retention can be difficult. We propose a simple and low-cost change in the way colleges use the ACT exam in their admission decisions that can greatly increase their ability to identify students at a high risk of under-performing and dropping out. Specifically, we show that only two of the four sub tests of the ACT, English and Mathematics, can effectively predict outcomes in college. This result is robust across various samples, specifications and outcome measures. We demonstrate that by eliminating the noise associated with the two non-predictive sub tests, student-college matches can be significantly improved.