To reduce the number of database scanning and reduce the burden of memory further, also to improve the efficiency of mining frequent itemsets better, an Apriori-based optimization algorithm (M-Apriori) is proposed. The method stores frequent itemsets state by constructing the frequent state matrix and store the relationship between the transaction and itemsets by constructing the Boolean matrix. The algorithm scans the database only once and generates the initial frequent state matrix and the Boolean matrix during the initialization phase. On this basis, all frequent itemsets can be found directly without scanning the database repeatedly. Experiments show that M-Apriori algorithm has better performance and efficiency compared with the Apriori algorithm.