Because the optimization algorithm of support vector machine ( SVM) falls into local optimum easily and has many control parameters, a SVM optimized by surface-simplex swarm evolution (SSSE) algorithm is proposed and the classification of Motor imagery EEG signals is studied. The fuzzy entropy and AR model parameters of MI EEG signals were extracted as input features, and then SSSE is applied to parameters optimization of SVM to classify MI EEG signals. In the test experiment, which classified the 2003 international brain-computer interface (BCI) competition Data sets Ⅲ and the 2008 BCI competition Data sets 2b by left-hand and right-hand. The results showed that the average classification accuracy and Kappa value of the proposed method were 82. 47% and 0. 88 respectively. SSSE reduced the control parameters and effectively avoided the particles falling into the local optimum. The validity of this method in the classification of MI EEG signals was verified.