In the circumstances of agile manufacturing,changes and uncertainties greatly increase in the production planning and scheduling.This paper deals with a new multi-agent architecture and a hybrid method for the dynamic scheduling of agile manufacturing system.Firstly it proposes a distributed multi-agent based manufacturing structure,which has characteristics of self-determination and distribution grounded on multi-agent as well as control and harmony grounded on hierarchical structure or dynamic logical unit (,which is temporarily formed for a specific task).Then it studies a hybrid job shop scheduling methods that combined with fuzzy programming and frizzy Contract Net Protocol.We classify the job shop scheduling into static planning and dynamic scheduling.In usual status,when batch orders arrived,scheduling agents model the fuzzy scheduling program at first,considered the fuzzy conditions such as fuzzy machining time or fuzzy delivery date; and then cooperated with algorithm agents to select an appropriate algorithm (in our research,we take GA for example);consequently,they solve the program to bring forth a whole scheme; according to the scheme,the tasks are assigned to resource agents following the hierarchical structure.While in emergency status,machine breaking down or new tasks inserted for example,it should adjust original scheme to adapt the new conditions.Thus modulating scheduling is relative small on scale but high random,so it feasible for dynamic and real-time scheduling where the negotiation between the agents for the scheduling of tasks is accomplished by the use of the fuzzy Contract-Net Protocol.Fuzzy decisions are made by the fuzzy logic modules,which embedded in the processes of biddings.The core of the module is a fuzzy reasoning machine that nmning on principle of fuzzy relationship synthesis.In a word,the hybrid method has both virtues of precision of static planning and flexibility of fuzzy Contract Net Protocol.