Abstract: The problem of optimally approximating linear systems is solved by the multiagent genetic algorithm (MAGA). In MAGA, each possible solution, a real-valued vector in the search space, is considered as an agent, and all agents live in a latticelike environment, with each agent being fixed at a lattice-point. In order to increase energies, they compete or cooperate with their neighbors, and they can also use knowledge. Therefore, four evolutionary operators are designed for simulating the intelligent behaviors of agents, such as competition, cooperation, self-learning and so on. Making use of these agent-agent interactions, MAGA realizes minimizing the objective function value. At the same time, a search-space expansion scheme is adopted to find the regions where the optimal parameters locate. In experiments, two linear systems, a stable one and an unstable one, are used to test the performance of MAGA, and a comparison is made between MAGA and two recent algorithms. The results show that MAGA can find high quality approximate models with low computational cost.