Celular automata are one of many alternative models of computation. Massive parallelism and the posibillity to describe their local behaviour in a simple way are of particular interest. This thesis describes a different way of representing the local transfer function of cellular automata, which is particulary suitable for use in genetic algorithm. This representation is based on simple model, mirroring the way instruction based register machines operate. The aim of this publication is to analyze and assess applicability of proposed method.