Sorting is the key factor of Fisherface recognition.This article proposes a Fisherface sorting method based on weight.First, in the lower-dimensional subspace of PCA, different weights are assigned for the examples of the same sample according to their distance to mean value, and then recalculated mean value.Then we can seek for eigenfnces space, and endow every eigenvector different weights in line with eigenvalue.In this way we restructure scatter matrix in the class and between the classes.This method improves fisher discriminant function.At last make out the distance between test examples and weighted subspaces, and make assortment recognition on distance.Proved by the experiment of YALE faces database, this method is better than traditional Fisherface.