It is a key issue in P300 Brain-computer interface (BCI)system that how to extract effective features of P300 potential from raw electroencephalography (EEG)data and get better classification performance. In this paper, we proposed a P300 potential detection method based on deep belief network (DBN)which can extract useful feature information from raw data without data preprocessing. First we converted the original EEG data as input data according to the time and spatial features, then exacted the features by the constructed DBN model, and finally realized the classification of P300 potential by the Softmax classifier. The results showed that the DBN has a good feature learning ability for the P300 potential and has achieved the better detection results.