The cost estimation now is a vital concern of every manufacturing firm
, and many methods are available for cost estimation. In this paper, w
e build an novel architecture of multiagent system with each agent rep
resenting one kind of cost estimation method (distributed multi-agent
system). The system first classifies the user's requests,then makes th
e expert agents to cooperate and compete to fulfill the task. The mode
l is based on a fuzzy classification of cost estimation methods and a
dynamic optimization of structure to make the agents to learn from the
past (based on agent reputation). In this way, the amount of informat
ion exchange is considerably reduced.