Abstract: This paper evaluated the risk of reservoirs collapse caused by nature-flood-based dam overtopping using a Bayesian network. Probabilities of single and cascade reservoirs collapse were calculated for reservoirs in Sichuan Province. The analysis demonstrates that Bayesian network is intuitive and user-friendly to investigate the risk of reservoirs collapse under multiple risk sources. A same conditional dam overtopping probability magnitude of 10^-6 is derived, indicating a low risk of single reservoir dam overtopping. However, the conditional probability of dam overtopping is over 0.8 when the water storage level is higher than normal, meaning there is a high probability of dam overtopping under this condition.