Abstract: Maximal frequent itemsets mining is a fundamental and important problem in many data mining applications. Since the MaxMiner algorithm first introduced the enumeration tree for MFI mining in 1998, there have been several proposed methods using depth-first search to improve performance. Here presented is DFMfi, a new depth-first search algorithm for mining maximal frequent itemsets. DFMfi adopts bitmap data format, several popular prune techniques which prune the search space efficiently, and local maximal frequent itemsets for superset checking quickly. Experimental comparison with the previous work indicates that it accelerates the generation of maximal frequent itemsets obviously, thus reducing CPU time.