Abstract: In view of the information gain model defects in the text classification, this article puts forward a text classification algorithm based on the grey relation and information gain. Firstly, we improved a method of x2 statistics in sort feature selection in order to express text. In this way, we improve the precision of the class center vector. Secondly, according to the IG model weights the low frequency words too bigger, we put forward an improved weighted method basing on frequency and position. Lastly, we put forward a new way in text similarity calculation in order to improve the shortcomings of the similarity calculation model that based on distance. Subsequent text categorization test shows that this paper puts forward an improved IG method and enhances the text classification efficiency.