Objective For the artifact signal of electroencephalography (EEG) in brain computer interface (BCI),this paper presents an artifact removal method based on second-order blind identification (SOBI) in blind source separation. Methods Firstly,joint approximate diagonalization and data whitening are utilized for multiple-channel. Meanwhile the mixing matrix is calculated and these EEG signals are decomposed into an equal number of independent component. Then,some independent components containing artifacts need to be set zero based on experience. And the remaining components are reversely projected and reconstructed with the mixing matrix to obtain EEG signals that artifacts are removed. Finally,the proposed method is tested from two aspects including the processing time and recognition accuracy based on three sets of experimental data. Results The proposed method has better performance than the commonly used independent component analysis (ICA). The processing time of one trial is shortened by 169.1ms,177.0ms and 230.8ms,and the recognition accuracy is increased by 3.3%,5 % and 10%. Conclusions The proposed SOBI can quickly and effectively remove artifact signals,which may lay the foundation for online processing of EEG in BCI.