Many organizations use anti-spam filters to detect and quarantine unsolicited commercial e-mail. However, these filters often flag legitimate messages as spam, causing problems such as delayed communication, missed business opportunities, and disrupted workflows. Since failure to control spam is often seen as a security lapse, an important question for managers is how to determine the optimal filtration level for a firm's anti-spam filters. To our knowledge, no systematic way for making such a determination has yet been developed, although analogous techniques relating to the Receiver Operating Characteristic (ROC) curve have been discussed in the signal detection and medical fields. In this regard, we first apply the ROC curve within the context of filtering spam. Furthermore, this research proposes a new and direct method that finds the optimal threshold level without using the ROC curve. A closed-form solution of an optimal filtration level is derived from the model for a given anti-spam filter and a user's perceived costs. This optimization will be applicable in many areas in which ROC curves are used. Based on the results, implications for managers and implications for theories are discussed.