Abstract: For the BCI (brain-computer interface) to classify the different imagined movements of both left and right hands, the method of CSP (common spatial pattern) was used to extract the features of BCI 2003 competitive dataset. Then, the CSP based on a sliding time window was used to filter the EEG(electroencephalogram) data from the electrodes C3, Cz and C4, with the SVM (support vector machine) used as a classifier of the features. As a result, the highest accuracy of classification is 82.86% with the best classification time point 4.09 sec, maximum mutual information (MI) 0.47 bit and maximum MI steepness 0.431 bit/s. Compared to the results of BCI 2003 competition, the method as above can provide greatly improved maximum MI steepness, thus verifying that the method is more adaptable to what are required by the online BCI system.