Mining user preferences plays a critical role in many important applications such as customer relationship management, product and personalized service recommendation. Although of great potential, to the best of our knowledge, the problem of mining user preferences from positive and negative examples has not been explored before. In this paper, we identify and model the problem systematically. Our theoretical problem analysis indicates that mining preferences from positive and negative examples is challenging. We develop a greedy algorithm called POUPMA and show the effectiveness and the efficiency of the algorithm using synthetic data sets.