We combine satellite-based pollution data and test scores from over 10,000 U.S. school districts to estimate the relationship between air pollution and test scores. To deal with potential endogeneity we instrument for air quality using: (1) year-to-year coal production variation; and (2) a shift-share instrument that interacts fuel shares used for nearby power production with national growth rates. We find that each one-unit increase in particulate pollution reduces test scores by 0.02 standard deviations. Our findings indicate that declines in particulate pollution exposure raised test scores and reduced the black-white test score gap by 0.06 and 0.01 standard deviations, respectively.