As current text classification depends on vector space model and document frequency lacks binary class- ification, a method based on class space model of difference frequency is presented in this paper. The method breaks the constraint on vector space model, and selects feature with difference frequency improved on document frequency, thus realizes the function of binary Classification. The experiment shows that improved method is effective. Three evaluation parameters, including Precision, Recall and F1, are im- proved in classification result, and classification precision is better. In addition, the method is worth learning in binary Classification of other areas.