We consider the problem of finding a commonly agreed upon diagnosis for errors observed in a system monitored by a number of different expert agents. Each agent is assumed to have its own specialized (expert) view on the system and collectively, the agents have to agree on one or more diagnoses based on their views. Reaching an agreement is complicated by the two factors: (i) different specialisms need not distinguish the same fault modes of a component and (ii) knowledge of different specialisms need not be correct in some cases. This paper analyzes these problems and presents protocols that enable the agents to deal with these issues.