The set of frequent closed patterns determines exactly the complete set of all frequent patterns and is usually much smaller than the laster. Yet mining frequent closed patterns remains to be a memory and time consuming task. This paper tries to develop an efficient algorithm to solve this problem. The compound frequent item set tree is employed to organize the set of frequent patterns, which consumes much less memory than other structures. The tree is grown quickly by integrating depth first and breadth first search strategies, opportunistically choosing between two different structures to represent projected transaction subsets, and heuristically deciding to build unfiltered pseudo or filtered projections. Efficient pruning methods are used to reduce the search space. The balance of the efficiency and scalability of tree growth and pruning maximizes the performance. The experimental results show that the algorithm is a factor of five to three orders of magnitude more time efficient than several recently proposed algorithms, and is also the most scalable one. It can be used in the discovery of non-redundant association rules, sequence analysis, and many other data mining problems.