Classification rules induction is a central problem addressed by machine learning and data mining. Rough sets theory is an important tool for data classification. Traditional rough sets approach, however, pursuits the fully correct or certain classification rules without considering other factors such as uncertain class labeling, importance of examples, as well as the uncertainty of the final rules. A generalized rough sets model, GRS, is proposed and a classification rules induction approach based on GRS is suggested. Our approach extends the variable precision rough sets model and attempts to reduce the inuence of noise by considering the importance of each training example and handling the uncertain class labels. The final classification rules are also measured with the uncertainty factor.