Market basket analysis is an important data mining application for finding correlations between purchasing items in transactional databases. Previous works show that considering constraints which users may concerned with into the mining process can effectively reduce the number of patterns and get more promising information. In this study, we extend the RFM analysis into the mining process to measure the importance of frequent patterns. In RFM analysis, a customer to be recognized as valuable if his/her purchasing records are recent, frequent, and having high amount of money. Follow the same concept of RFM analysis, we first define the RFM-patterns. The RFM-patterns we discovered are not only frequently occurred but also recently bought and having a higher percentage of revenue. After that, we propose a tree structure, named RFMP-tree, to compress and store entire transactional database, and a pattern growth- based algorithm, called RFMP-growth, is developed to discover all RFM-patterns from RFMP-tree. In experimental evaluation, the results show that the algorithm can both significantly reduce the number of discovered patterns and efficiently find the RFM-patterns.