The rules of classification of the group of N independent observations into one of k normal populations are considered. In the case when parameters are not known, Bayes estimators of the discriminant scores are suggested.