This project was divided into three parts. By storing a large data set in the database system, the data can be efficiently queried to analyze the customer base with reference to size and density, distribution, and vital statistics. First, the current customer portfolio was profiled in terms of demographics, net present value, and transactional behavior. Then the data mining techniques were applied to build empirical models. Here the main technique is the k-means, an algorithm of cluster analysis. Our goal was to find the group that made the most use of the reward programs and the group that was the most profitable for the company. After comparing the characteristics of these two groups, it was found that they were somewhat poorly matched. That means that the current reward program might have some problems, because the more profitable customers were not rewarded more. Second, the lagged regression analysis was used to explore the cause-effect relationship between spending and redemption. This information helped to some extent to judge ?the price of loyalty?. The results showed that there was some correlation between them, and it also provided estimated parameters in the regression models. Last, the current rewards scheme was evaluated and several possible schemes were also come up with. By making some reasonable assumptions and running the cost-benefit analysis, a modified scheme was recommended and it was showed to contribute better revenue for the company and to benefit customers as well.