Abstract: This paper proposes an improved algorithm for mining frequent itemsets based on the projection and sorting location index table of frequent itemsets. The data structure of "projection" is set up by scan database once, and find the frequent 1-itemsets directly and obtain frequent 2-itemsets by inner product operation. Then, the location index table of the higher order itemsets is established, find the frequent itemsets by leaping searching and linking. Through the experimental analysis, the efficiency of finding frequent itemsets is greatly improved.