Multi-contrast MRI offers us images with complementary diagnostic information. Despite the dramatic difference in contrast, multi-contrast images often share highly correlated structure information. A deep learning (DL) based strategy is proposed to denoise multi-contrast MR images with flexible noise-levels using residual U-Net. This method utilizes the structural similarities across contrasts by simultaneously denoising multiple contrasts while existing single-contrast MRI denoising methods neglect the analogous structure information. The proposed method outperforms BM3D in terms of better noise reduction and details preservation. More importantly, we introduce a noise-level map that can be manually set to fit the different noise levels.