Abstract: Currently, feature extraction and classification are carried out separately in electroencephalogram (EEG) recognition. To simplify the identification process and improve the classification results, we propose a novel method that applies a deep belief network(DBN) to EEG recognition. Deep belief networks consist of several layers of restricted Boltzmann machines (RBMs) and one layer of a back-propagation network (BP). In the process of EEG recognition, RBMs are used to extract features from the EEG data to ensure that the most effective feature vectors are obtained. These feature vectors are then classified by the BP network. We collected motor imagery EEG data with an Emotiv EEG acquisition instrument, and used them to design the experiment. The experimental results show that DBNs have a strong ability for feature learning from raw EEG data. The recognition rate of motor imagery EEG data based on a DBN is better than that from a support vector machine (SVM). The proposed method simplifies the EEG recognition process, improves the recognition rate, and provides a novel technique for the brain computer interface(BCI) recognition of EEG signals.