The theory of modern immunology provides a novel idea to study network intrusion detection and defence system. Inspired information processing in biology immune system is a highly parallel and distributed intelligent computation which has learning, memory, and associative retrieval capabilities. The architecture of multi-agent in depth defence based on immune principle is proposed. The agents of intrusion detection detect all intrusion which passes by the agent, including known and unknown. The information of new intrusion, which gotten from current monitored network is encapsulated and sent to the other network as vaccine by mobile agents. So the other network can prevent the same intrusion. Intrusion packets are prevented from gateway of intrusion source by response agent. The experimental results show that the new model actualizes an active and distributed prevention policy than that of the traditional passive intrusion prevention systems.