Abstarct: In recent years, the large number of firms which offer similar services, has helped customers to choose from among several firms baxsed on their needs and enjoy their services. This diversification is clearly seen in many industries such as insurances, telecommunications, internet service providers and cable TV networks and caused the strategy of “keeping customers and avoiding to lose them to the rivals” to become one of the most important managerial strategies. Cutting the relationship with the current service provider is called “customer churn”. To lose a profitable customer means that he is being attracted by rival service provider and the cost of attracting a new customer is much more than the cost of keeping him (Moser, 2002). From the economic and risk management point of view, identifying the customers whose churn will incur great loss is crucial. The existing information about customers, including loyal and churned customers, is a baxse for predicting the future customer behavior. If we could predict the probability of churning customers by examining the characteristics of customers, we would be able to bring the churning rate of customers to the minimum by exerting preemptive activities. There are a number of methods for predicting the customer churn, such as logistic regression, neural networks, decision tree and the Rough Set Theory. In this thesis, we introduce the aforementioned models and investigate the accuracy and efficiency of them on the case study using different evaluation criteria. The results showed that, these criteria suggest different models for predicting customer churn.