In　the　field　of　data　mining　and　pattern　recognition,　classification　is　a　very　important　core　technology.　This　paper　present　two　kinds　of　improved　classification　algorithm.　Using　the　improved　Naive　Bayes　(NB)　and　KNN　algorithm　structure　classifier　to　filter　normal　mail　and　spam.　Improved　NB　algorithm　can　dynamically　adjust　the　threshold　k,　reduces　the　mail　mistake　rate.　Center　vector　method　is　introduced　into　the　similarity　calculation　formula　of　KNN,　better　reflect　the　interrelation　between　the　text　and　categories.　Finally,　improvedNB　algorithm　and　KNN　algorithm　make　comparison　and　analysis,　it　is　concluded　that　the　effective　experimental　results.