A stable anisotropic diffusion model for edge-preservlng image smoothing is presented. By comparison with existing diffusion models~ this model has two significant new features. The first is that the diffusion model has the ability to adjust the diffusion coefficient and to regulate diffusion orientation simultaneously. The second is that the diffusion model being nonlinear in essence is linearized along time coordinates which makes the model more stable and thus the complexity of computation is reduced greatly. We use an Additive operator splitting scheme (AOS) that is an efficient and stable numerical method to implement the edge-preservlng image smoothing. The experimental results show that the model proposed in this paper is not only stable and efficient but can preserve image edge.