The goal of this work is to propose a complete data mining system for the solution of customer retention problems, commonly found in many industries. Such a solution encompasses the accurate identification among huge amounts of data of those consumers who would most likely end their relationship with the firm, based on their historical behavior and individual profile. Acting upon the intelligence provided by a precise customer classification, incentives and retention actions should be put into practice to prevent or minimize the losses of valuable clients to competitors. Throughout the data mining process designed here, great care was given to the preparation and representation of the data and to input selection methods, in an effort to optimize the performance of the classification models. Various different classification techniques have been tested, with the objective of finding the one best suited for the task at hand: to pinpoint those customers who present clear risks of abandoning the analyzed company. Among the studied models were neural networks, decision trees, genetic algorithms, neuro-fuzzy systems and SVMs (Support Vector Machines). As a case study, the issue of churn (loss of customer to a competitor) in the Brazilian wireless telecommunications was tackled, due to the availability of data. A detailed study was made, identifying the causes, consequences and details of the business problem. As a conclusion, the great impact of the implementation of the proposed system in retention strategies of wireless carriers is evaluated, under the view of the profitability that would be generated by its use.