Abstract. This paper deals with the synchronous implementation of situated Multi-Agent Systems (MAS) in order to have no execution bias and to ease their programming on massively parallel computing devices. For this purpose we investigate the translation of discreteMAS into Cellular Automata (CA). Contrarily to the sequential scheduling generally used in MAS simulations, CA are a model for massively parallel computing where the updating of the components is synchronous. However, CA expressiveness is limited and not always adapted to all types of modeling situations, especially when independent entities move in space. After illustrating these issues on a simple example, we propose a generic method to translate a discrete MAS into a CA, called a transactional CA. Our approach consists in using the influence-reaction model to perform this translation.