ing at the problems in the face recognition algorithm based on Weber features that the directional information is not made full use and the extracted information is also insufficient, we propose a novel face recognition method based on multi-directional Weber gradient histograms. On the basis of original differential excitation, the neighborhood pixel gradient is increased, and the improved differential excitation and Weber gradient features are extracted. The improved differential excitation and Weber direction are quantized, and the two-dimensional histograms are extracted in blocks, which are further converted into one-dimensional histogram features. The histogram features are extracted along the Weber direction. Two features are connected to form a compound feature and simultaneously the nearest neighbor classifier is used for classifying. The experiments on different face databases show that the proposed method has not only a good recognition effect, but also a relatively strong robustness to illumination, expression and partial occlusion.