Customer Relationship Management System (CRM) has accumulated massive customer transaction data. Effective customer segmentation by analyzing transaction data can contribute to marketing strategy designing. However, the state-of-the-art researches are defective such as the uncertain number of clusters and the low accuracy. In this paper, a novel customer segmentation model, AP-GKAs, is proposed. First, factor analysis extracts customer feature based on multi-indicator RFM model. Then, affinity propagation (AP) determines the number of customer clusters. Finally, the improved genetic K-means algorithm (GKAs) is used to increase clustering accuracy. The experimental results showed that the AP-GKAs has higher segmentation performance in comparison to other typical methods.