Spam is considered a significant security problem for computer users everywhere. Spammers exploit a variety of tricks to conceal parts of messages that can be used to identify spam. This paper presents an email feature extraction technique based on artificial immune systems. The method extracts a relatively small set of features that can be used as inputs to a classification model. A Backpropagation neural network is employed as the spam detection model. The performance evaluation against a standard spam collection and reference systems shows that the proposed approach performs well compared to other systems with large sets of features, rules, or external evidences. The detection rate of the best system in the study is 92.4%, with 1% and 13.8% of false positive and false negative rates, respectively.