Presently, a variety of rule-based classification methods in e-mail filtering obtain good results. In the training of e-mail filtering, the training set has the notion that some e-mail messages will be sent to the hazy category. Extracting these e-mails from training set will have a noticeable increase in the results of classification. Therefore, a clustering-based filtering method is proposed in this paper. The common features of the hazy-category email include cluster the training set. Experiments demonstrate that the method has better performance on the appraisal standard than that of a simple rule-based classification algorithm.