Objective To improve the training efficiency of brain computer interface(BCI)training based on motor imagery，the paper designs a training system including parameter setting，electroencephalography(EEG)acquisition，feature extraction，classification，feedback of results and the foundation of classifier model. Methods Firstly,the training system acquires the EEG signals of motor imagery by EEG acquisition software，and then transmits data to MATLAB with TCP/IP protocol.So the feature extraction and classification are realized in the MATLAB. After that,the training system feedbacks the recognition results with the progress bar at the same time，and allows subjects to adjust their status in the training. Finally the system generates an effective classification model by choosing a proper feedback mode in a short time. Results The characteristics of the system are convenient intervention，powerful function and friendly interface. A BCI online system is established to test the training system and an effective classification model can be completed in a relatively short period of time. Conclusions The training system can improve the recognition rate and shorten the training time for BCI application systems，which lays the foundations for BCI application systems.