Mining algorithm of FP-Tree is one of the most effective mining algorithms in association rule mining. It must produce large amounts of the candidate set and scan database repeatedly, but it generates conditional FPTree recursively in the process of mining frequent pattern and wastes the storage space greatly using the common tree's memory structure. It proposed an algorithm for mining frequent patterns by constructing reverse FP-tree with a binary tree storage structure. In mining process, it mines left son tree recursively, but gets frequent pattern not by generating conditional FP-Tree, so it greatly reduces the storage space and running time. The experiments show that the improved algorithm can realize effective mining on the time and space.