In this paper, the purpose of association mining is to find the valuable relationships between data sets. The prerequisite of it is to find the frequent itemset first. In view of the existing problems in the present frequent itemset mining, this paper puts forward that data sets should be clustered first, and then the algorithm of frequent itemset mining be applied to every cluster. In this way, algorithm of mining can be easier, and the memory consumption be decreased as well when the super frequent itemset is found. Besides this, the overlooked frequent itemset with lower general support but higher cluster can be found.