Abstract: To improve the evaluation accuracy of flexural capacity of corroded reinforced concrete (RC) structures, a method for the model parameter updating and prediction of flexural capacity is proposed based on the improved particle filter (PF) algorithm. The proposed model comprehensively considers the geometric size of corroded RC structures, steel cross-sectional area and mechanical properties, concrete strength, bond performance and other factors. A large number of particles are generated to represent the uncertainty of model parameters during the degradation process of flexural capacity. The PF algorithm is improved from the perspective of selecting different proposal density functions to solve the problem of particle degradation in traditional PF algorithm. The PF, the extended particle filter (EPF) and the unscented particle filter (UPF) algorithm are employed to estimate and update the model parameters, which can effectively predict the flexural capacity of corroded RC structures. The results show that the flexural capacity of RC beams decreases gradually with the increase of steel corrosion loss. The prediction method of the flexural capacity of corroded RC structures based on the improved PF algorithm considers the updating of model parameters, which makes the prediction results closer to the reality. The improved PF algorithm based on EKF and UKF can effectively constrain the degradation of the particle, and the prediction accuracy is better than that of the PF algorithm. The prediction accuracy of the flexural capacity of corroded RC structures increases with the increase of training data and particle number.