In this paper document classification methods using multinomial na¨ıve Bayes are improved in a number of ways. We use the value weighting method, a new fine-grained weighting method, to calculate the weights of the feature values. Our experiments show that the proposed approach outperforms other state-of-the-art methods.