Mining　frequent　closed　itemsets　from　data　streams　has　been　studied　extensively.　Algorithm　MOMENT　and　its　modified　algorithm　A-MOMENT　were　regarded　as　typical　methods.　Both　of　them　depend　on　a　data　structure　named　CET.　This　paper　designs　a　new　data　structure　FULL-CET　and　proposes　a　new　mining　frequent　closed　itemsets　algorithm　MFCIDS　based　on　landmark　window.　Differing　entirely　from　traditional　methods　which　find　new　frequent　itemsets　through　union　operations　on　existed　frequent　itemsets,　MFCIDS　records　the　support　of　each　closed　frequent　itemset　to　maintain　all　frequent　closed　itemsets　through　intersection　operations　on　nodes　appearing　actually　in　transactions.　Experimental　results　show　that　MFCIDS　performs　better　than　MOMENT　and　its　modified　algorithm　A-MOMENT　on　efficiency　and　scalability.