The paper considers a multi-agent model of a computer networks security system, which is composed of particular autonomous knowledge-based agents, distributed over the hosts of the computer network to be protected and cooperating to make integrated consistent decisions. The paper is focused on an architecture, implementation and simulation of a case study aiming at exploration distinctions and potential advantages of using such an architecture for the computer network protection. The paper describes the conceptual model and architecture of the particular specialized agents and the system on a whole as well as implementation technology. Simulation scenario, input traffic model and peculiarities of the distributed security system operation are described. The major attention is paid to the intrusion detection task and agents interactions during detection of an attack against the computer network. The advantages of the proposed model of a computer networks security system are discussed.