During　the　process　of　personalized　recommendation,　some　items　evaluated　by　users　are　performed　by　accident,　in　other　words,　they　have　little　correlation　with　users＇　real　preferences.　These　irrelevant　items　are　equal　to　noise　data,　and　often　interfere　with　the　effectiveness　of　collaborative　filtering.　A　personalized　recommendation　algorithm　based　on　Associative　Sets　is　proposed　in　this　paper　to　solve　this　problem.　It　uses　frequent　itemsets　to　filter　out　noise　data,　and　makes　recommendations　according　to　users＇　real　preferences,　so　as　to　enhance　the　accuracy　of　recommending　results.　Test　results　have　proved　the　superiority　of　this　algorithm.