Personalization　is　a　major　characteristic　of　Web　intelligent　system　for　the　future.　To　provide　personalized　commodity　recommendation,　this　paper　presents　a　new　recommendation　mechanism　based　on　multilevel　customer　model,　which　formalizes　the　recommendation　process　as　the　learning　of　multilevel　customer　model,　the　generating　of　recommendation　set　and　personalized　filter.　The　mechanism　captures　customer＇s　needs　from　three　aspects:　shopping　demands,　preference　characteristics　and　purchasing　power,　thus　enhances　the　degree　of　personalization　in　recommendation　and　improves　the　effectiveness　of　recommendation.