The performance of support vector machine (SVM) heavily relies on the setting of the hyperparameters. An evolutionary Monte Carlo (EMC) method was i n troduced for SVM hyperparameters optimization. The target function was cross v a lidation accuracy, a population of samples was simulated in parallel. An improve d mutation operation was designed for accelerating the mixing of the Markov chai ns and improving the searching efficiency. It makes the algorithm possess strong global searching ability in the initial stages and refined searching ability in the posterior stages. The optimal SVM was applied to bearing fault diagnosis. T he experimental results show that the proposed method has good optimization effi ciency and classification ability, and improves the accuracy of the fault recogn ition.