This paper investigates the impact of extending free school lunch to all students, regardless of income, on academic performance in New York City middle schools. Using a difference-in-differences design and unique longitudinal, student-level data, we derive credibly causal estimates of the impacts of "Universal Free Meals" (UFM) on test scores in English Language Arts (ELA) and mathematics, and participation in school lunch. We find UFM increases academic performance by as much as 0.083 standard deviations in math and 0.059 in ELA for non-poor students, with smaller, statistically significant effects of 0.032 and 0.027 standard deviations in math and ELA for poor students. Further, UFM increases participation in school lunch by roughly 11.0 percentage points for non-poor students and 5.4 percentage points for poor students. We then investigate the academic effects of school lunch participation per se, using UFM as an instrumental variable. Results indicate that increases in school lunch participation improve academic performance for both poor and non-poor students; an additional lunch every two weeks increases test scores by roughly 0.08 standard deviations in math and 0.07 standard deviations in ELA. Finally, we explore potential unintended consequences for student weight outcomes, finding no evidence that UFM increases the probability that students are obese or overweight. We also find no evidence of increases in average body mass index (BMI). Instead, we find some evidence that participation in school lunch improves weight outcomes for non-poor students. Results are robust to an array of alternative specifications and assumptions about the sample.