Abstract: This paper presents a mixed method of reducing the text features which based on the feature selection and the feature extraction.Firstly, we carry out the first selection through improving on the odds radio.The text is expressed with the matrix vector which composes of sort attributes.Then, we use an improved scatter difference to extract the features again.In the condition of the least information lost, we have completed the text feature reduction twice.The result which based on the Chinese text categorization shows that this method has a better precision in the text categorization.