Many layout design problems can be considered as constraint satisfaction problems. However it is not easy to obtain the optimal solution in many cases uniquely. In this paper, we tried to use genetic algorithm for furniture arrangement in a room according to user's planning intention. The orientation and position of furniture are coded into bit strings of genes. After processing of crossover breeding and random mutation, higher fitness solutions are selected. This process is executed iteratively. The key point of this research is not to find unique optimal solution, but to find some of suitably fitted solutions according to user specified constraints. For this purpose, we discussed man machine interface concretely in user's planning process. It was clarified that in the selection process of genes based on constraint fitness, elite-keeping selection is suitable to improve fitness gradually and selection by roulette has possibility to derive a different type layout.