This paper presents a new adaptive image denoising scheme by combining the nonsubsampled contourlet transform (NSCT) and Stein’s unbiased risk estimation (SURE). The original image is first decomposed by using NSCT. Then the mean square error (EMS) is estimated based on Stein’s unbiased risk estimation. The noises of each decomposed subband are reduced by using the linear adaptive threshold function, which can be constructed based on the EMS. Finally, the denoised image is obtained after reconstructing the processed subbands. Experiments and comparisons on both standard images and natural images show that the proposed scheme can remove image noises effectively and outperforms the current schemes in regard of both the peak signal-to-noise-ratio (PSNR) and the edge preservation ability.