In　order　to　solve　such　problems　as　over-scale　of　the　grid　portal　resources　management,　heavy-load　of　handing　with　large-scale　querying　and　processing,　low-satisfaction　of　the　users　who　need　access　to　get　the　desired　the　resources,　we　present　an　improved　grid　portal　recommendation　architecture　in　this　paper.　The　proposed　architecture　in　combination　with　a　collaborative　filtering　algorithm　allows　for　the　main　features　of　the　grid　portal.　In　addition,　the　paper　implements　the　architecture　efficiently　through　a　prototype　portal　in　which　both　action　layer　and　render　layer　are　designed　for　collaborative　filtering.　Lastly　the　experiment　results　show　that　the　adoption　of　this　architecture　can　be　used　to　achieve　grid　portal　personalized　collaborative　filtering　recommendation　function　and　guarantee　accuracy　and　quality　of　personalized　recommendation.　©　2008　IEEE.