The concept of similarity is fundamental to pattern recognition and associative memories. In recent years many mathematical models with various neural network struc- tures and different learning algorithms have been proposed. Generally in these mo- dels, the neural network emphasizes the association within the stored objects, and such models are quite effective under the assumption that the stored objects signifi- cantly differ from one another (are independent)1. Unfortunately, in practice, the objects are usually not independent, moreover, the differences among the patterns are often very small, and recognition methods based on the similar features fail. Having a number of similar objects, for example human faces, with many features identical, what features do we use to recognize a particular individual?