Abstract: Traditional spam filtering techniques based on email` s content used to implement text-related machine learning and classification. Nevertheless, the uncertainty of message`s content causes a performance bottle for machine`s discrimination and classification. Aiming at the deficiencies of traditional spam filtering methods, this study brings forward a vector supported classifier model based on the features of spammer`s behaviours. The evaluation results for real spam testing set show that the spam classifier based on features of spammer`s behaviour is capable of discriminating email`s type with pretty high accuracy. Meanwhile, the classifier model is of robust performance on noise data.