Text classification is the foundation of information retrieval and text mining. Naive Bayes can be applied to text classification as it is simple and efficient classification. But the two assumption about Naive Bayes algorithm that its attribute independence is equal to its attribute importance are not always consistent with the reality, which also affects the classification results. It is a difficult problem to disapprove the assumptions and improve the classification effect. According to the characteristics of text classification, based on text mutual information theory, a Term Weighted Naive Bayes text classification method based on mutual information is proposed, which uses the mutual information method to weight the feature in different class. The effect of two assumptions on classification is partially eliminated. The effectiveness of the proposed method is verified by the simulation experiment on the UCI KDD data set.