Network-level access control policies are often specified by various people (network, application, and security administrators), and this may result in conflicts or suboptimal policies. We have defined a new formal model for policy representation that is independent of the actual enforcement elements, along with a procedure that allows the easy identification and removal of inconsistencies and anomalies. Additionally, the policy can be translated to the model used by the target access control element to prepare it for actual deployment. In particular, we show that every policy can be translated into one that uses the “First Matching Rule” resolution strategy. Our policy model and optimization procedure have been implemented in a tool that experimentally demonstrates its applicability to real-life cases.