Abstract: In existing reinforced concrete buildings, due to the particularity of the environment in which the structure is located, predictions based only on empirical data according to similar historical records of existing buildings may be misleading. Bayes’ theorem can provide a comprehensive consideration of empirical data and detection data, the prediction model can be dynamically updated according to the detection data. Since a large number of random samples need to be extracted during the Bayesian analysis, the Kriging adaptive approximation model is introduced. By comparing to the calculation results of the finite element model, it is shown that the approximate model can improve the calculation efficiency and at the same time ensure a high accuracy. Applying Bayesian theory to the prediction model of stiffness degradation coefficient of reinforced concrete frame structure, it is found that the introduction of test data has a significant impact on the posterior distribution, and the prediction of structural performance highly depends on the update of information. The prediction model based on Bayesian theory can provide a prediction model that conforms to the actual situation of the current project.