Unsolicited e-mail became a major source of frustration for Internet users; in addition to that, these undesired e-mails are causing companies to lose billions of dollars each year in lost workers productivity. Present paper proposes a method to automatically classify new incoming e-mails as good or bad. The method uses a machine learning based engine that learns the keywords for the two categories from a set of training e-mails. Then, these keywords are used to predict the label of e-mails in the test set. Proposed method correctly categorizes 98.2% of good e-mails and 74% of bad e-mails.