Abstract: Along with the widespread concern of spam problem, at present, there are spam filtering system about the problem of semantic imperfection and spam filter low effect in the multi-send spam. This paper proposes a model of spam filtering which based on Latent Semantic Analysis (LSA) and Message-Digest algorithm 5 (MD5). By making use of the LSA marks the latent feature phrase in the spam, a semantic analysis is introduced into the spam filtering technique; the "e-mail fingerprint" of multi-send spam is born with MD5 on the LSA analytical foundation, the problem of filtering technique's low effect in the multi-send spam is resolved with this kind of method. We design a spam filtering system based on this model. This system is evaluated with an optional dataset. The results obtained are compared with Naïve Bayes algorithm filter experiment results. The experiments show the expected results, and the feasibility and advantage of the new spam filtering method is validated.