Abstract: A new method of decreasing the dimension of feature vector by using the theory of pattern aggregation(PA) is presented.The method connected Kohonen network acquire better result of text categorization.The Kohonen network is applied to realize text classifying,and apply supervising method to network training.Therefore,the speed and the precision of classifying are improved.However,to the text vector of high dimension,the speed of classifying is still very slow using Kohonen network.Even the result of classifying cannot be acquired.The new method establishes vector space model of term weight by the theory of PA,which enhances the function of the words from the viewpoint of categorization effect,and decreases the dimension of vector by eliminating redundant features.Therefore the new method advances the speed and the precision of text categorization largely,and the method has better generalization ability,which is approved by the experimentation.