We have introduced methods for obtaining discriminant functions by learning. The clues used so far were learning patterns themselves, and the information about the probability distribution of learning patterns was assumed to be unknown. In this chapter, we first describe a method for obtaining discriminant functions assuming that the information about the probability distribution of learning patterns is known. Next, guidelines for designing discriminant functions are introduced from a different perspective. Next, we explain that it is extremely important to understand the relationship between the dimensionality of the feature space and the number of learning patterns in order to design a classifier, along with the reasons for this. Finally, we describe how to evaluate the designed classifier based on the error probability. Specific methods for setting up learning and test patterns will be introduced.