This thesis proposes four novel methods based on multiresolution subbands for face recognition. The first method is based on Multiple Classifiers System. In this method, 2-level subbands fusion is proposed. The second method focuses on selecting the least redundant subbands. Mutual information is used to measure this redundancy. This leads to a set of subbands which are the most independent. However, using only independent subbands may not cover all information. Thus, in the third method, the subband selection is based on information coverage instead of the subband dependency as in the second method. Information coverage is measured according to mutual information. In the second and the third methods, the new subband selection depends on the previously selected subbands. The result of greedy incremental selection is not guaranteed to be optimal. Thus, in the fourth method, we use information coverage criterion and reselect a new set of subbands for each desired number of subbands.