Learning from examples algorithms is a traditional method for data mining. Usually the examples undergoing learning are partitioned into positive and negative sets. Discriminative rules are discovered by generalizing process on positive sets and specializing on negative sets. These algorithms require an explicitly specified negative example set and only effect in an ideal environment, which doesn't exist because of the uncertainty in management domain. How to discover knowledge from this kind of data sets is a difficult problem and the proceeded research work is limited. The main contribution of this paper is to propose an approach for discovering knowledge from uncertain data with incomplete information based on the refined rough set model of information systems.