As an important research area in pattern recognition and computation geometry, the symmetry of image has many applications in object recognition, visual inspection and shape representation etc. A novel approach that uses distributed active agents to extract the basic reflectional symmetry axes in gray images is presented in this paper. It detects, groups and links the prominent local symmetry axes by simulating the behaviors of the agents such as inhabit, evolve, diffusion and death in local image environme.It can extract the basic reflectional symmetry axes of the arbitrary gray images and is suitable for parallel implementation.The experimental results on the natural images show that this approach is sfficient.