The research of the relationship between eyemovement and human activity has become a hot spot in the field of the human-computer interaction. In order to recognize the eye-movement patterns related to reading tasks, an algorithm based on electrooculogram(EOG) has been proposed in this paper. In the proposed approach, wavelet packet decomposition was firstly used to smooth EOG signals during reading tasks, and then, according to the predefined coding strategy, the smoothed EOG waveform is translated into a coding string to represent current reading state. Finally, Levenshtein distance is employed to measure the similarity of the template string and the output string. The experiments under the experimental paradigms of autonomy-reading and stimulus-reading and reading with rest state have been carried out. The mean accurate has reached 90%. The experimental results indicate that the proposed algorithm can effectively recognize human reading activity state.