In order to promote the performance of EEG classification based on multi-task motor imagery (MI), we used common spatial pattern (CSP) as the feature extraction method, and we extracted the features under two conditions, with one "One versus One" and the other "One versus Rest". Then, as for the different feature extraction methods, we presented different classification methods based on support vector machine (SVM) according to the different input features. The final classification results showed that the mean Kappa of "One versus One" classification method based on decision value is much higher than that of voting rule, and a little higher than that of "One versus Rest" classification method.