Abstract: An image nonlinear filter based on Partial Differential Equations (PDE) has good performance, but it consumes large time and resource. Cellular Neural Networks (CNN) can depict the spatial discrete PDE model, and by means of an CNN analog chip, CNN can solve PDE efficiently. A nonlinear filter based on CNN-PDE was studied, and for selecting the diffusion coefficient properly a noise-estimate technique was presented by means of local operation only. The test result showed that this noise-estimate technique offered a comparatively accurate measure of different noise levels. Simulations of artificial noise images showed that this CNN-PDE nonlinear filter would suppress noise and preserve image edge simultaneously. It is feasible and effective to realize the PDE image process technique by CNN.