Abstract： In order to solve the slow transmission rate and low classification accuracy,we proposed a multi-classification method for electroencephalogram(EEG)signals by using one vs rest filter bank common spatial pattern(OVR-FBCSP)and sparse embeddings(SE).For reducing the complexity of multi-task feature extraction and improving the classification efficiency,the OVR-FBCSP was used to extract EEG feature.Then,the corresponding label matrix was embedded in low dimension,the SE model was constructed,and the embedding matrix of the training and test data were calculated respectively.Finally,k-nearest neighbor(kNN)classification was performed for training and test data in the embedding space.The experiment was tested on the BCI Competition IV-2a open data set and compared with other classification methods.Experimental results show that the proposed method has higher classification accuracy and shorter analysis time.