The general procedure of image denoising is solving an inverse problem with some statistic prior information as the regularization. We propose a novel noise removal method for Magnetic Resonance (MR) images based on the Fuzzy Gaussian Mixture Model (FGMM) Clustering of image patches. In this method, the FGMM, which is an extension of the Gaussian Mixture Model (GMM), has been trained using overlapping clean patches randomly selected from the image database firstly. Then the objective function, which is the sum of the image corruption model and the prior model, is constructed. We optimize the objective using the “Half Quadratic Splitting” method and obtain an expression of iteration to restore the whole image. Finally, we use the proposed method to denoise Magnetic Resonance (MR) images. The experimental results show that the proposed method achieves good performance in MR image denoising.