Mining frequent closed itemsets provides complete and non-redundant result for the analysis of frequent pattern. Most of the previous studies adopted the FP-tree based conditional FP-tree generation and candidate itemsets generation-and-test approaches. However, those techniques are still costly, especially when there exists prolific and/or long itemsets. This paper redesigns FP-tree structure and proposes a novel algorithm based on it. This algorithm not only avoids building conditional FP-tree but also can get frequent closed itemsets directly without candidate itemsets generation. The experimental results show the advantage and improvement of these strategies.