Along with the increasing size of database and the reduction of support threshold, the number of frequent patterns will grow exponentially, and the time and space efficiency of the FP-growth algorithm will greatly reduce. The cause of low efficiency was analyzed, and according to the analysis, a lattice-based algorithm for fast mining frequent itemsets (LFP-growth) was presented. The proposed algorithm divided a large lattice into many sub-lattices by using equivalence relation. Through iterativing decomposition of sublattices, frequent itemset mining in lattice was transformed into frequent itemsets mining in a union set of multiple sublattices. Experiments have shown that the time and space performance of LFP-growth algorithm is superior to that of FP-growth algorithm in mining large database.