Two image denoising approaches based on wavelet neural network (WNN) optimized by particle swarm optimization (PSO) are proposed. The noisy image is filtered by the modified median filtering (MMF). Feature values are extracted based on the MMF and then normalized in order to avoid data scattering. In approach 1, WNN is used to tell those uncorrupted but filtered by MMF and then the pixels are restored to their original values while other pixels will retain. In approach 2, WNN distinguishes the corrupted pixels and then these pixels are replaced by MMF results while other pixels retain. WNN can be seen as a classifier to distinguish the corrupted or uncorrupted pixels from others in both approaches. PSO is adopted to optimize and train the WNN for its low requirements and easy employment. Experiments have shown that in terms of peak signal-to-noise ratio (PSNR) and subjective image quality, both proposed approaches are superior to traditional median filtering. Two images denoising approaches based on wavelet neural network (WNN) optimized by particle swarm optimization (PSO) are proposed. The noisy image is filtered by the modified median filtering (MMF). Feature values ​​are extracted based on the MMF and then normalized in order To avoid data scattering. In approach 1, WNN is used to tell those uncorrupted but filtered by MMF and then the pixels are restored to their original values ​​while other pixels will retain. In approach 2, WNN distinguishes the corrupted pixels and then these pixels are WNN can be seen as a classifier to distinguish the corrupted or uncorrupted pixels from others in both approaches. PSO is adapted to train and the WNN for its low requirements and easy employment. Experiments have shown that in terms of peak signal-to-noise ratio (PSNR) and subjective image quality, both proposed approaches are superior to traditional median filtering.