Aiming at the channel selection and classification issue of EEG signals, a motor imagery EEG classification model based on group sparse Bayesian logistic regression (gsBLR) is proposed, which can simultaneously accomplish channel selection and classification. Firstly, spatial filtering and bandpass filtering are performed on the multichannel signals to reduce the influence of volume conduction effect. Secondly, the time domain, frequency domain and timefrequency domain features with discriminant information are extracted for each channel signal, and feature fusion is performed. Finally, the gsBLR method is used for channel selection and classification. The model parameters are automatically estimated from the training data under the Bayesian learning framework, which avoids cumbersome and timeconsuming crossvalidation process. Experiments were carried out on two public BCI competition datasets and selfcollected dataset, and the highest average classification accuracies of 8163%, 8497% and 7647% were achieved, respectively. Compared with other methods, the proposed method achieves better classification accuracy and fewer number of channels. At the same time, the selected channels are more compatible with the neurophysiological background.