Summary: Mining frequent closed itemsets is a fundamental and important issue in many data mining applications. A new depth-first search algorithm for mining frequent closed itemsets called depth-first search for frequent closed itemsets (DFFCI) is proposed which could reduce the number of candidate itemsets and the cost of support counting. DFFCI processes the dataset information stored by the improved Compressed Frequent Pattern tree (CFP-Tree) into the partition matrix, and improves the efficiency of support counting by using the binary vector logic operation. Global 2-itemset pruning based on support pre-counting and local extension pruning is used to prune the search space effectively. The experimental results show that DFFCI outperforms other depth-first search algorithms.