A new method of speaker recognition based on Biomimetic Pattern Recognition and improved nearest neighbor method has been proposed. This new model based on "matter cognition" instead of "matter classification" in traditional statistical Pattern Recognition, which is closer to the function of human being. We experiment with the BPR by using improved nearest neighbor, which act through covering the high dimensional geometrical distribution of the sample class in the feature space. It is exciting that the recognition rate can achieve 100% in text-dependent test, and about 98.57% with 8.75% rejection rate and 1.43% error rate in text-independent test which shows that this method is efficient and practical.