This paper proposes a novel feature extraction method for face recognition in the wavelet domain called wavelet projection entropy (WPE). First, the projection entropy features from each wavelet subband are computed along the vertical and horizontal direction after the division. Then information fusion scheme is applied to integrate results obtained from each subband. Experiments show that WPE can extract the meaningful information from the wavelet domain. Meanwhile the decision level fusion achieves the best recognition rate among the three common information fusion methods. The proposed algorithms are validated on ORL and Yale face database for different pose and expression changes analysis. Detailed comparisons with previous published results are provided and it shows that our proposed algorithm performs very well.