The cross-entropy method which is used in solving the optimal probability mass function(PMF） of power generation system is based on the iteration of a single PMF, resulting in large number of state samplers must be needed and decreasing the efficiency of the calculation of reliability indexes consequently. We proposed a cross-entropy method based on the iteration of a mixture of many PMFs with different weights and take samplers to calculate the reliability indexes based on the mixture. With the proposed method, the convergence of the indexes will be faster while the accuracy is satisfactory, the samplers needed in the reliability evaluation of high-reliable system will be much less. Computational results demonstrate the superiority of the proposed method.