In this paper, a multi-agent system is employed to model urban systems. The supposed urban system is composed of three urban agents (residence, business, and service areas). Agents interact each other economically and pay land rents and moving costs. Agents who can not adapt to the urban system die. Agents who survive in the system move or create new agents according to the information of environment, internal states, and their own rules coded on genomes. When a new agent is created, his parent's genome is followed to the new one. Using this model, urban land-use pattern formation is simulated with three different distance restrictions. In the case of slight restriction, patterns are concentric and moving areas are wide. In the case of strong restriction, pattern become concentric and moving areas become shorter. Patterns are compared with the land-use of Osaka area using fractal dimension and its similarity is confirmed.