Policies exhibit conflicts much as features exhibit interaction. Since policies are defined by end users, the combinatorial problems involved in detecting conflicts are substantially worse than for detecting feature interactions. A new, ontology-driven method is defined for automatically identifying potential conflicts among policies. This relies on domain knowledge to annotate policy actions with their effects. Conflict filtering is performed offline, but supports conflict detection and resolution online. The technique has been implemented in the RECAP tool (Rigorously Evaluated Conflicts Among Policies). Subject to user guidance, this tool filters conflicting pairs of actions and automatically generates resolutions. The approach is generic, but is illustrated with the APPEL policy language for call control. The technique has improved the scalability of conflict handling, and has reduced the effort required of the previous manual approach.