For motion-related brain-computer interfaces（BCIs）,some artifacts may be imported due to the setting of unnecessary electrodes,which is unfavorable to the feature extraction and classification.In order to solve this problem,an electrode selection method based on the common spatial pattern（CSP） and the binary particle swarm optimization（BPSO） is proposed,which extracts features with high discriminant ability and limits the number of electrodes at the same time.Moreover,in order to accelerate the convergence of BPSO,the generation probability of initial electrode combinations is adjusted according to the weights of CSP filters.Experimental results show that high classification accuracies can be achieved only with few electrodes.