This paper presents a Multi Agent System (MAS) design with distributed intelligence for fault location and isolation in power distribution systems with the presence of Distributed Generation Sources (DGS). In the proposed MAS, agents all over the feeder communicate with their neighbors and use the local differential current information to locate the faulty zone and isolate the fault. Agents update their local knowledge by exchanging their voltage and current phasor data with their neighbors and monitoring the local current. The distributed generation penetration is considered to be up to 50 percent. The multi-agent models are simulated in Matlab® Simulink using user defined s-functions and the power system is modeled using the Simulink Simpower toolbox. The proposed method has been tested on a model of an existing Mon Power company circuit. Both faulted zone and fault type have been successfully identified.