Using General-Purpose computing on Graphics Processing Units (GPGPU) in Multi-Agent Based Simulation (MABS) appears to be very promising as it allows to use the massively parallel architecture of the GPU (Graphics Processing Units) to improve the scalability of MABS models. However, this technology relies on a highly specialized architecture, implying a very specific programming approach. So, it turns out that MAS models need to be adapted if they want to benefit from GPU power. Contrary to some recent research works that propose to hide GPU programming to ease the use of GPGPU, we present in this paper a methodology for mode-ling and implementing MABS using GPU programming. The idea is to be able to consider any kind of MABS rather than addressing a limited number of cases. This methodology defines the iterative process to be followed to transform and adapt a model so that it takes advantage of the GPU power without hiding the underlying technology. We experiment this methodology on two MABS models to test its feasibility and highlight the advantages and limits of this approach.