Abstract: In the case of random simulation of small probability events of component faults in power system, the traditional non sequential Monte Carlo simulation method for power system reliability evaluation has such defects as needing large sampling number and long simulation time. By combining antithetic variable sampling method with cross entropy importance sampling method, an improved sampling method, which is suitable for power system reliability evaluation, is therefore proposed. An approximate function with zero variance probability, whose optimal parameters of component are obtained using cross entropy importance sampling density function of component, is constructed first. Then the variances of reliability indices are further reduced through antithetic variable sampling according to the obtained optimal parameters. The sampling efficiency is thus improved and advantages of the two aforementioned methods are complemented. The proposed method, the traditional random sampling method, the antithetic variable sampling method and the cross entropy importance sampling method are respectively applied to reliability evaluation of IEEE-RTS and modified IEEE-RTS test system. A comparison of the results shows that, compared with the other three methods, the proposed method has higher sampling efficiency is improved, and its advantage is more obvious in rarer events.