It’s important for mobile operators to recommend new services. Traditional method is sending advertising messages to all mobile users. But most of users who are not interested in these services treat the messages as Spam. This paper presents a method to find potential customers who are likely to accept the services. This method searchs the maximum frequent itemsets which indicate potential customers’ features from a large data set of users’ information, then find potential customers from those maximum frequent itemsets by using a bayesian network classifier. Experimental results demonstrate this method can select users with higher accuracy.