Abstract: For the current structural reliability analysis of multidimensional variables, the response surface methods (RSM) with the uniform design are widely used. To deal with the limitation of fitting the models in the utilization of the original Least Squares (LS) regression, a new approach based on the traditional method has been proposed for improvement. The method combines Partial Least Squares (PLS) with a uniform design to fit the response surface under the condition of small samples data with strong correlation in order to calculate structural reliability. The results of several examples shows that the Partial Least Squares method proposed in this paper can be used, especially in the high dimensional problem, to analyze the structural reliability, and higher accuracy can be obtained by comparing it with the original Least Squares technique.