Recent controversies have highlighted the importance of local school district governance, but little empirical evidence exists evaluating the quality of district policy makers or policies. In this paper, we take a novel approach to assessing school district decision making. We posit a model of rational decision making under uncertainty that emphasizes districts learning over time. We test the predictions from the model using data on a set of highly visible and consequential decisions facing school district leaders--the choice of learning mode during the 2020-21 school year. We find that district behavior is consistent with a Bayesian learning process in several key respects. Districts respond on the margin to health risks: All else equal, a marginal increase in new COVID-19 cases reduces the probability that a district offers in-person instruction the next week. This negative response is magnified when the district was in-person the prior week and attenuates in magnitude over the school year, suggesting that districts learn from experience about the effect of in-person learning on disease transmission in schools. We also find evidence that districts are influenced by the learning mode decisions of peer districts, but not their peers' experiences with in-person instruction and disease transmission, which implies that some important frictions exist.