In this paper, the feature extraction method based on the WPT (wavelet packet transformation) is proposed according to ERD/ERS phenomenon in the motor imagery EEG signals. The method takes the data of the 2005 braincomputer interface competition as the process object. Firstly, EEG signals are preprocessed. Then the average energy difference of wavelet coefficients in the specific frequency bands between C3 and C4 channels extracted with WPT is taken as feature vector. Finally, BP neural network is adopted to classify motor imagery EEG signals of right and left hands.