The recognition of unusual customers is one of the most important applications in data mining. Using an example from telecom trade, This work presents a kind of classification and regression tree algorithm to recognize unusual customers in telecommunication trade databases (TCART for short). The TCART includes the following process: maximal classification tree construction, regression pruning and optimal classification tree selection. During the procedure of classification tree growing, we propose improved classification approaches. Finally, the TCART algorithm is validated by large numbers of customer data in the Zhejiang telecom database. The experimental results indicate that the TCART algorithm has the virtues of robustness and effectiveness.