Aiming at the prominent and serious leakage problem of retaining structures of deep foundation pits in water-rich sand layers, a dynamic risk analysis method for deep foundation pit leakage based on the Bayesian network is proposed to adapt to the change of leakage risk over time during excavation and improve the objectivity of evaluation. Taking the leakage disease of deep foundation pits in water-rich sand layers in Hangzhou as the background, a risk analysis index system covering stratigraphy, supply, structure, and deformation was established. The static Bayesian network model was constructed using the Bayesian network and the Leaky-Max hypothesis to analyze the sensitivity of leakage risk probability and the risk factors as well as indices of loss. The results of engineering case analysis show that the dynamic loss, probability and risk level of leakage during the excavation of deep foundation pits in water-rich sand layers obtained by this method are consistent with the actual engineering situation, which verifies the rationality of the method that can be used to predict the leakage risk of retaining structures of deep foundation pits in water-rich sand layers.