A method of classification based on wavelet packet and Sample Entropy was proposed to improve the correct classification rates of mental task EEG signal. Firstly, With the help of wavelet package, the original EEG signals are decomposed and then the energy feature related to frequency bands and time were extracted from the original EEGsignals.Thirdly, the energy and sample entropy are used as a vector and the vector was used by BP Neural network to classify EEG-based motor imagery tasks.The results showed that the new method based on the combinatorial feature of the energy and sample entropy which were extracted during imagining left right hands movement with the proposed method, had better classification results than the traditional method based on solely single feature. This paper provides new ideas and methods of feature extraction and classification of different mental tasks for BCI.