Abstract: Naive Bayes (NB) is a simple and efficient classification algorithm, and it is widely used in spam filtering, but because of the independence between the attributes of the hypothesis which has to some extent affected in classification effect. To solve this problem, the FOA-NB algorithm is proposed which is an improved NB algorithm. The algorithm of NB algorithm and Fruit fly optimization algorithm (FOA) combination, according to the different feature attributes of the influence degree of the classification given different weights, to optimize the weights by FOA and get the global optimal feature weight vector, the algorithm in the NB algorithm retains the advantage of simple and efficient at the same time, by optimization of the weights to obtain attributes which have more decision-making, so as to improve the spam filtering correct rate and recall rate. Through the simulation experiment with NB algorithm, Weighted Bayesian (WB), the results show that the FOA-NB algorithm makes the spam filtering effect has been improved significantly, and the correct rate and recall rate are improved, and the increase of about 5%.