This article focused on the realization of the parallelization of Naive Bayes. When it comes to large-scal data or multi-attributes, the traditional singal node algorithm has a low efficiency,or even is unable to host large-scale computing. All of these make the traditional algorithm cannot fit the need to deal with massive data. Therefore, based on the basic theory of Naive Bayes and the framework of MapReduce, this paper proposed a parallelization method of Naive Bayes, which is efficient and cheap.At the end, it is proved by experiments that this method can effectively improve the efficiency of the algorithm so as to meet the need of peoople to deal with massive data.