Aiming at the problem of traditional methods for damage identification being difficult to distinguish multi-damage states and the reliability of prediction results, a method for damage identification was proposed based on autoregressive (AR) model and Gaussian process (GP).It was shown that this method uses AR model to fit acceleration response data of a structure, and for the first time, parameters L₁ and L₂ are  introduced to characterize the information about damage location and damage level; based on residual and coefficient of AR model, damage sensitivity features for locating damage position and identifying damage degree are established, respectively; GP’s classification algorithm is combined with its regression algorithm to realize multi-damage location and probability output of damage degree.The effectiveness of the proposed method was verified with a numerical simulation example of a reinforced concrete model arch, and the identification results based on AR model’s residual and coefficient were analyzed contrastively.The results showed that the proposed method can identify multi-damage states and output prediction results with probabilistic significance, which is helpful to judge the reliability of the results and realize damage early warning; the identification accuracy and reliability of damage sensitivity feature based on residual are higher with better anti-noise performance; in the case of 10% noise pollution, the relative error mean and discrete coefficient mean of the identification results are only 6.52% and 0.19, respectively.