Abstract: In this paper, we describe a new pattern recognition approach for the four-tone classification which is important for Putonghua speech recognition. In this method, four measurements on the pitch frequency track are made on the base of a large number of statistical experiments. And under the assumption that the measured parameters satisly multidimensional Gaussian distribution (which is shown to be rational), a distance function for each tone type is derived according to the rule of minimum probability of classification error, so that optimum decision results in statistical meaning are achieved. The proposed method is able to provide very satisfactory results as shown by the experiments of speaker-independent four-tone class ification.