[Purposes] Excavation of large cross-section tunnels is prone to pose safety hazards to the stability of superstructures. In order to accurately predict the risk level of surface buildings, a risk assessment model of a large cross-section tunnel underneath buildings was constructed. [Methods] Based on the static Bayesian network, a risk assessment model was established, which included four first-level indicators including geology, tunnels, building structures, and relationship between tunnels and building locations, as well as 14 second-level indicators. Through the classification of risk status and the fuzziness of expert language, the prior risk probability values were obtained. By using the coefficient of variation (CoV) and arc spacing algorithm, the key risk factors affecting surface buildings and their strength of influence were obtained. Furthermore, the dynamic Bayesian network model was established, and the Genie software modules such as “Noisymax node” and “Strength of influence” were used to update the model reasoning results by analyzing the on-site monitoring data from the Baitashan Tunnel project underneath the Baihua Pavilion. Then, the risk change trend of surface buildings in the whole construction process was calculated. [Findings] The soil factor is the most critical risk factor, followed by the tunnel diameter, settlement rate of cave roof, convergence rate of surrounding area, cumulative settlement of cave roof, cumulative convergence of surrounding area, and other tunnel factors, with groundwater and poor geology at the lowest rank. In addition, groundwater, poor geology, and construction management are risk factors with the largest strength of influence. The risk change trend data obtained from the dynamic Bayesian network has an error of only 5.0% compared with the on-site monitoring data of construction.[Conclusions] The risk assessment method for large cross-section tunnels underneath buildings proposed in this paper can quantitatively analyze the key risk factors and their strength of influence. Combined with the engineering monitoring data, the dynamic prediction of the risk of surface buildings can be realized, which can provide some theoretical and practical guidance for similar projects.