Abstract: The whole structural space of random variables can be divided into reliable space and failure space. Based on the adaptive sampling skill in the Monte Carlo method, the sampling point in failure space can be selected efficiently. Therefore, the distributed parameters of reliable space and failure space can be determined. Then the performance function is restructured by means of Mahalanobis distance analysis method. Consequently, a new method is proposed to analyze structural reliability with small samples. It is of universality since it only uses the state of sampling point, the reliable or the failure. Numerical simulation experiments show that the new method is of higher precision with small samples.