COVID-19 led to school closures and emergency remote learning. This paper analyzes school closures during the pandemic using a unique data base. The determinants of the duration of school closures estimates were used to instrument school closures -- stringency of lockdown and vaccination -- and causally estimate the impact of duration on learning. It is estimated that for every week that schools were closed, learning levels declined by almost 1% of a standard deviation. This means that a 20 week closure, for example, would reduce learning outcomes by 0.20 standard deviation, almost one year of schooling.