Abstract. This paper presents an agent-based, high-level security system for user verification. The system verifies a user's identity by processing data from several low-lever sensors mounted at an entry point combined with knowledge of the user's past behavior. The data from a new entry are processed by several agents, which store the knowledge in an ontology. A single agent's classifications are integrated into the overall decision. The system successfully detects intruders, even when they optimally fake low-level sensors, e.g. fingerprints, and regular personnel under the influence of drugs.