This paper provides new evidence on the long- and medium-term impact of extreme weather events on education. Our focus is on Mongolia, where two extremely severe winters caused mass livestock mortality. We use household panel data with information on households' pre-shock location, combined with historic district-level livestock census data and climate data. Our econometric strategy exploits exogenous variation in shock intensity across space and time, using a difference-in-differences approach. Results indicate that individuals who experience the shock while of schooling age and living in severely affected districts are significantly less likely to complete mandatory education, both in the long- and medium-terms. The effects are driven by individuals from herding households, while no significant effects are found for individuals from non-herding households. This finding renders it unlikely that extreme winters affect education through school closures during extreme climatic conditions, to which all children were exposed. Moreover, there is no evidence for a differential impact of extreme weather events by gender. This suggests that the effects are not mainly channeled through increased child labor in herding, but rather they are related to reductions in household income.