The Shearlet multi-orientation features fusion and weighted histogram are proposed to overcome the disadvantage of Shearlet transform, which has data redundance in extracting features and cannot sparsely represent the global characters. First, Shearlet transform is used to extract the multi-orientation facial features. Then two coding methods are proposed to fuse the features from different directions of the same scale into a single feature, and the fused image is divided into a number of equal-sized nonoverlapping rectangular blocks, weighted fusion of each model using the Shannon entropy theory. Many experiments have been done on the ORL, FERET and YALE face database, which fully proved that this method has more advantages in terms of recognition than the traditional Shearlet.