Abstract: KNN is one of the classical algorithms in text categorization.The number of training samples and the density is the primary bottleneck on the algorithm.A reasonable method for reducing the amount of training data can improve the efficiency of classification.This paper proposes an improved KNN model basing on clustering.Firstly, by clustering the samples into clusters, we remove some samples from training set basing on the distance in order to save computing cost.Secondly, take into account the category distribution we bring up a better weighting method in order to overcome the defect that the bigger class of training samples dominated in KNN.The result of test shows that the improved KNN classification algorithm improves the efficiency of its classification.