This paper presents a new Monte-Carlo method for reliability evaluation of power systems. The estimation of reliability indices is based on the stochastic simulation of system states and state transition processes. Such a choice provides the convenience to analyze the working and repair processes in power system operation, and the possibility to estimate failure frequency and unsupplied energy in the system. In the technique of simulation, the theory of conditional expectation and conditional distribution is employed. A method of using the state duration expectation to replace the duration sample is, for the first time, proposed in the paper, which reduces the estimator variances signif icantly. Doing this, the effects on the successive state transition processes are taken into account by the recursive conditional CDFs. A set of recursive convolution formulas is presented in the paper. With the discription of its computer method, the paper also discribes the Monte-Carlo method's application to a 4-unit generation system with scheduled maintenance. The results are compared with those of the direct simulation and the comparison shows that the model and method proposed are correct and efficient.