The research paper focuses on developing a holonic multi-agent system (HMAS) to assist the central company during the formation of manufacturing networks. The manufacturing network paradigm gives a central company control over which group of manufacturers will participate on its customers' orders. The central company therefore has flexibility over the lead time, the production cost and the conformance reliability of its manufacturing networks. In order to simulate the impact of a central company's operation strategy on the performance of manufacturing networks, this research proposes an HMAS framework for network formation simulation and an HMAS simulation platform. The framework depicts the integration of the holonic manufacturing paradigm, agent-based technology, artificial neural network, ontology and discrete event simulation. Resulting from the development of the HMAS, an algorithm for manufacturer selection is also proposed. The simulation platform was tested, verified and validated with data from an industrial case study. The proposed algorithm is also compared with the algorithm used by an actual central company. The impact of the operation strategies of a central company on the performance of manufacturing networks is also presented. The HMAS platform and decision algorithm exhibit significant potential in manufacturing networks formation.