The response surface method is regarded as an effective way to solve the nonlinearity and complexity problems of the seismic reliability of bridges，but it has problems in selecting both agent model types and training samples． In order to solve these problems，an improved sequential sampling response surface method is proposed based on the infill sampling criterion ( ISC) optimization method and the Kriging model． In the proposed method， by taking the Kriging model as the agency and by drawing on the Monte Carlo sampling technique，an ISC function，which includes the predictive values of unknown samples and the corresponding MSE，is established by making use of the excellent prediction performance of the Kriging model and its unique ability to estimate the mean square error ( MSE) ． In the iterative phase，by solving the ISC optimization problem，the local and global searches are conducted and then the subsequent training samples are generated sequentially． Finally，the proposed method is utilized to analyze the dynamic reliability of a suspension bridge under random seismic excitation． The results show that the proposed method is of high accuracy and high efficiency．