Abstract: In order to solve the problem that the brain computer interface (BCI) application need to analysis EEG signal rapidly and precisely , we proposes a method of EEG feature extraction and recognition based on convolutional neural network (CNN) and common spatial model (CSP). On the basis of the preprocessing of the original EEG signal, the corresponding feature matrix is obtained by CSP space transformation. Study on application of CNN feature matrix, the fully connection layer weights analysis on the convergence of the CNN network, according to the characteristics of e-learning definition CSP matrix feature selection criterion, obtained EEG efficient dimensional feature set F, computing feature set scale of F to construct CNN classifier.Experimental tests were carried out in BCI2005 IV a competition data set, obtained 88.3% recognition accuracy, compared to the traditional CSP such as sCSP algorithm and KLCSP method, the average recognition accuracy were improved by 3.2% and 2.4% of the results. In this paper, the method combines the time and space characteristic information, and introduces a new view of the secondary selection of the feature according to the result, which provides a new idea for the feature extraction of EEG.