Different denoising schemes show dissimilar types of artifacts. For example, certain transform-based denoising schemes could introduce artifacts in smooth regions while others eliminate texture regions. Using different schemes for denoising a noisy image, we can consider the denoising results as different estimates of the image. Through linear combination of the results, we minimize the l2 norm of the error to find the optimum coefficients in a least-square-error sense. We employ the wavelet transform, contourlet transform, and adaptive 2-D Wiener filtering as our denoising schemes. Then we apply the proposed method to the denoising results of the individual schemes. This approach eliminates most of the artifacts and achieves significant improvement in the PSNR values. We also propose averaging of the denoising results as a special case of linear combination and show that it yields near-optimal performance.