Abstract: In order to reduce the error rate in classification calculation of anti-spam filtering system, the automatic testing process of target email in the Bayesian anti-spam filtering system was analyzed, the definition of system cost was researched from two aspects of system filtering quality and user fault tolerance. Cost parameters were analyzed in the collection of different sample sets and attribute spaces, with disabling and enabling lemmatizer and stop-list. By adjusting the cost parameters, the results of the Bayesian filtering system in various assumptions were analyzed, the standard of system modeling was optimized, and system filtering quality was upgraded. The results prove that the scheme is feasible.