In sequential pattern recognition systems, the selection and ordering of effective features from a given set of feature measurements is an important problem. The purpose of this paper is to discuss the efficiency of the divergence and Chernoff's distance (particularly. Bhattacharyya's distance) as a criterion of feature selection and ordering. After these probabilistic measures were reviewed, a very simple multiclass pattern recognition was simulated as an example of the application of these. In this situation, to maximize the expected divergence and the expected Bhattacharyya's distance was adopted as a criterion of feature selection and ordering. The relations of these measures to the number of features and to the mean probability of misrecognition were obtained.