In this paper, an intelligent distributed multi-agent system (MAS) is proposed for the implementation of a novel optimization technique for distributed voltage regulation. The proposed MAS approach controls a large heavily-meshed distribution network which is grouped into small subnetworks using ϵ decomposition. The voltage regulation is accomplished by distributed generator (DG) agents, linear programming solver (LPS) agents, network violation detector (NVD) agents, and one ϵ decomposition agent. The LPS agent has an embedded control algorithm which optimizes DG generation within a subnetwork once the voltage at particular nodes exceeds the normal operational limits. The subnetworks and their control requirements are achieved through self-organization, which is the novelty of the research. Each intelligent agent has its own knowledge and reasoning logic to plan its own activities. The control actions are coordinated through agent communications within the subnetwork. The agent platform, Presage2, with improved autonomy and agent communication capability, has been used to develop the proposed MAS system and design the agents' behaviors.