Reliable, cost-efficient, and fast medical diagnosis is still a challenge in today's world. This paper presents a medical diagnosis system that combines the advantages of multi-agent system technologies and neural networks in order to realize a highly reliable, adaptive, scalable, flexible, and robust diagnosis system for diseases. The medical diagnosis system consists of a structured alliance of medical experts - realized by agents - that collaborate in order to provide a viable medical diagnosis. Each agent has a certain responsibility. The agents rely on a reactive pattern-based matching process. Their power is substantially augmented by standard kinds of neural networks (interactive neural nets and self organizing maps), which represent the diagnosis capacity of the agents.