In this paper, an improved KNN classification algorithm is proposed by using characteristics that the points distributed in the same category of sample collection are in close distance as an assistant to classify KNN algorithm. The way to deal with the k-nearest neighboring sample points is calculating the average distance between categories that the sample points belong to and the differences of unspecified sample points respectively. If the data calculated is greater than a certain threshold, delete this sample point from k-nearest neighboring samples, then determine the categories of unspecified sample points through majority voting. The improved KNN algorithm enhances the precision of classification and maintains the same time complexity as the traditional KNN algorithm.