A tree-based algorithm for mining of frequent patterns, TBA-FP, is proposed in this paper. The term, tree-based, has two meanings. First, a tree-based representation is used to compress the pattern information in the database. Second, the mining process is regarded as a process of constructing a frequent pattern tree in depth first order. Optimizing techniques, such as pseudo clipping, is discussed. Experiments show that TBA-FP is much more efficient and scalable than the classical algorithm, Apriori.