Abstract: Discussing the limitation of data source in conventional collaborative filtering recommendation (CFR), which is widely used in B2C E-commerce websites, an improved CFR based on customers' transaction data is put forward. The fundamentals of the improved CFR are described as follows. The three customers' behavioral indices including the recency (R), frequency (F), and monetary (M), are acquired from customers' historical records of shopping for various goods to evaluate their purchasing preferences, and then an integrated RFM (IRFM) matrix is formulated as the data source of CFR to recommend personalized goods for target customers, thus providing a high-quality recommendation to familiar customers so as to expand the application range of CFR.