The power system reliability evaluation methods are time-consuming and generate large errors. In order to solve these problems, a method for reliability evaluation of power systems is proposed based on the improved particle swarm optimization(PSO), the least squares support vector machine(LSSVM), and Monte Carlo simulation(MCS), where the improved PSO is used to optimize the parameters of SVM. The method can obtain more accurate LSSVM parameters through reasonable improvement of PSO, and a PSO-LSSVM model for classification of system state samples has been established. Using the PSO-LSSVM model to classify the system samples extracted by the MCS method, failure state samples and normal state samples were obtained with the PSO-LSSVM model. Reliability indices only for failure state samples were calculated and reliability evaluation results were output. This method was used to calculate the reliability index of the IEEE-RTS 79 system under different operation conditions, and the results show that the method improves the evaluation precision of the LSSVM-MCS method with a consistent amount of computation time.