Abstract: The Naïve Bayes classifier has been a popular classifier in many fields. However, it improves the prediction effciency at the expense of prediction accuracy because of the conditional independence assumption. In this paper, we first adopt the Rosenblatt transformation to obtain some mutually independent covariates and then apply the Naïve Bayes classifier. The mutual independence among features significantly increase the performance of the corresponding classifier. Meanwhile, in order to stabilize and simplify the Rosenblatt transformation, we propose using the PC algorithm to identify some conditional independence structures. We call the resulting classification model the Rosenblatt-Naïve Bayes model. This model takes the dependence among features automatically, and can maintain a simple model structure at the same time. We show the superiority of the Rosenblatt-Naïve Bayes model in terms of both prediction accuracy and model robustness through extensive simulation studies and three real data analyses.