A new Monte Carlo Method for system reliability evaluation is proposed in this paper. The repeated sampling of system states with relatively great probabilities is avoided in the simulation,thus the sampling efficiency is significantly improved. Besides,the negative correlations used in different trials can further reduce the estimated varinaces. The sparse matrix technique is employed to reduce the memory requirement and to save the computer time. Comparative studies show that the method proposed is of great value and the application to the IEEE reliability test system (power system) is successful.