When Gaussian noise and salt and pepper noise both exist in image, single mean filter or median filter turn out to be dissatisfactory. The characteristics of noises and dominance of filter algorithms were analyzed. A hybrid filter and fusion algorithm based on neural network was proposed. Firstly, probabilistic neural network was built to detect the salt and pepper noise and Gaussian noise and remove them respectively by median filter and mean filter algorithm. Then trained radial basis function neural network was built to fuse the two kinds of different filtering image. The ideal fusion image was output finally. The results by Matlab simulation experiments showed that the proposed algorithm can effectively remove mixed noise and preserve image edges and details very well. It is an effective method of image denoising.