Spam, also known as Unsolicited Commercial Email (UCE), has been an increasingly annoying problem to individuals and organizations. Most of prior research formulated spam filtering as a classical text categorization task, in which training examples must include both spam emails (positive examples) and legitimate mails (negatives). However, in many spam filtering scenarios, obtaining legitimate emails for training purpose is more difficult than collecting spam and unclassified emails. Hence, it would be more appropriate to construct a classification model for spam filtering from positive (i.e., spam emails) and unlabeled instances only; i.e., training a spam filter without any legitimate emails as negative training examples. Several single-class learning techniques that include PNB and PEBL have been proposed in the literature. However, they incur fundamental limitations when applying to spam filtering. In this study, we propose and develop an ensemble approach, referred to as E2, to address the limitations of PNB and PEBL. Specifically, we follow the two-stage framework of PEBL and extend each stage with an ensemble strategy. Our empirical evaluation results on two spam-filtering corpora suggest that the proposed E2 technique exhibits more stable and reliable performance than its benchmark techniques (i.e., PNB and PEBL).