Abstract: To raise the efficiency of data mining, a graph-based algorithm for mining frequent closed itemsets, called GFCG (graph-based frequent closed itemset generation) was proposed. In this algorithm, the bit vector technique is used to construct a directed graph to represent the frequent relationship between items, and frequent closed itemsets are generated recursively from this graph. As a result, the GFCG scans the database for only two times, and does not generate candidate sets. Furthermore, the concept of an expanded frequent itemset is introduced to greatly decrease the searching range for adjusting whether a frequent itemset is closed or not. In addition, this algorithm was tested by using one actual and two synthetical databases. The testing result shows that compared with the A-close and CLOSET algorithms, the proposed algorithm has good speed and scale-up properties.