To reduce the computational variance in the Monte Carlo method and accelerate the reliability evaluation of power system， a hybrid algorithm based on optimal sampling and selective analysis is proposed. Compared with the conventional Monte Carlo method， the new algorithm has a pre-sampling process so as to obtain the optimal probability density function and the variance projection. By this algorithm， the element states of analog variables can be sampled using the obtained optimal probability density function and the analytical variables can be determined according to the variance projection. The calculating results of the IEEE-RTS system show that the proposed algorithm can improve the efficiency of both the sampling and analytical calculations; the calculating variance is effectively reduced as well. This algorithm can also be applied to the reliability evaluation of various power systems， such as generation system， transmission system， composite system and distribution system， etc.