We study the distributional effects of remote learning using a novel approach combining preference data from a conjoint survey with administrative records. Experimentally derived preferences account for selection into remote learning and treatment effect heterogeneity. We validate the approach using random variation from school choice lotteries. On average, remote learning reduced reading and math achievement, but children whose parents showed strongest demand experienced positive effects. Parental concerns about bullying strongly predict demand, and remote learning consistently reduced bullying, partly offsetting learning losses. These results suggest that students who sort into post-pandemic remote learning may benefit from its expansion.