International audienceCurrently enterprises are organising themselves in 'new' structures, alternatively called enterprise networks, virtual enterprises, supply chains. We are interested in enterprise networks in which decision and information systems are distributed. We are searching for a global performance while controlling the architecture with local bilaterally negotiated co-decisions. We propose an architecture based on multi-agent system and the associated protocols, described by statecharts, to ensure as much as possible a global coherence and the convergence of the decision process. We propose models and tools in order to validate our approach by simulation of the multi-agent system. We employ our model on a particular problem in which enterprises are searching complementary resources (stocks, production capacities, components and/or transport capacities) in order to deal with unexpected orders different from the planned forecasting. The simulation considers parameters of the architecture systems in order to define how many unexpected orders must be treated in parallel the goal is to find a good equilibrium between minimisation of answer delay and maximisation of efficiency