Segmenting customers based on transaction data contributes to better understanding and characterizing customers, and has drawn a great deal of attention in literature of various fields. Data mining literature has provided various clustering algorithms for customer segmentation, and distance measure plays an important role in many approaches. However, most distance measures are based on co-occurrence of items, and pay few attention to the sales volume or quantities of items in transactions. In this paper, the probability density of items is employed to gather the description information of transactions and calculate the distance between transactions. Based on distinguishing the difference between similarity measures for transactions and customers, set distance is employed to evaluate the similarity between customers. The whole process is introduced as a framework to reach the target of segmenting customers.