The American Academy of Pediatrics recommends that middle and high schools start at 8:30 or later, yet fewer than 15% of schools follow this recommendation. Estimating the benefit of a later start time has proven difficult as any school’s start time is not determined randomly. Using quasi-experimental variation embedded in a school assignment algorithm, we measure the causal impact of school starting times on student learning. Our instrumental variables strategy leverages exogenous variation arising from tie-breakers in the Deferred Acceptance assignment mechanism. We use data on Wake County Public Schools entering middle schoolers who may apply to magnet schools with start times ranging from 7:30am to 8:30am. For children whose counterfactual school starts the day at 7:30am, attending a magnet school that starts at 8:30am is associated with a large and statistically significant gain in reading and math test scores. These results suggest that later school start times can substantially improve student learning.