Influenced by environmental factors, the performance evolution process of a reinforced concrete structural member is actually a stochastic process. It is not easy to predict the load-bearing probability distribution of a corroded reinforced concrete structural member by using traditional theoretical physical models due to simplifications and hypothesis while data-driven methods can do it well. But the data-driven methods are heavily dependent on the quantity and quality of inspected data. It still remains a challenge to integrate limited data to predict the load-bearing probability and quantitatively characterize the uncertainty of a corroded reinforced concrete structural member. This paper introduces a Bayesian updating method based on Markov Chain Monte Carlo (MCMC) method to integrate physical and data-driven models, improving the prediction of the normal section flexural bearing capacity of a corroded reinforced concrete beam. Initially, a data chain that matches the theoretical probability distribution for corrosion rate of longitudinal steel bars is generated, which is then input into the physical prediction model to obtain the prior distribution of stiffness. By combining the inspected deflection data and prior information, the corrosion rate is updated through stiffness by using Bayesian method. Finally, the posterior distribution of the corrosion rate is input into the physical model to obtain the posterior distribution of flexural bearing capacity. The results demonstrate that the Bayesian method can be effectively used to combine the inspection information and prior knowledge to revise the uncertain material mechanic parameters, thus leading to a more accurate predict results, which shows the potential of Bayesian methods in the assessment of existing concrete structures.