In this study, we examine how learning about disaster risk affects asset pricing in an endowment economy. We extend the literature on rare disasters by allowing for two sources of uncertainty: (1) the lack of historical data results in unknown parameters for the disaster process, and (2) the disaster takes time to unfold and is not directly observable. The model generates time variation in the risk premium through Bayesian updating of agents' beliefs regarding the likelihood and severity of disaster realization. The model accounts for the level and volatility of U.S. equity returns and generates predictability in returns.