This paper examines the impact of urban violence on the standardized test scores of public school students in Medellin. I use the school-level variation in exposure to local homicides for years 2004-2013 in a model that includes school and year fixed effects, allowing me to control for the endogeneity of violence. I find that each additional homicide per year occurring within 500 meters of a school reduces student achievement by slightly over 0.01 standard deviations on a variety of tested academic subjects. For an average school, this implies dropping from the 50th percentile to the 47th percentile in a typical violent period. The effect does not appear to be driven by bias from student migration, and evidence from differential effects estimates is more consistent with supply-side channels. Examining the causal pathways of the effect, the impact of local violence is shown to induce higher levels of teacher turnover, although this may actually improve the average qualifications of teachers in the school in the long run.