High-resolution　magnetic　resonance　images　can　display　richer　texture　details　and　facilitate　doctors＇　diagnosis　of　the　disease.　However,　limited　by　hardware　equipment,　the　resolution　of　clinical　magnetic　resonance　images　is　close　to　the　physical　limit.　Many　studies　use　super-resolution　reconstruction　to　enhance　magnetic　resonance　images,　but　they　ignore　the　image　noise　due　to　voltage　fluctuations　and　patient　jitter.　Direct　super-resolution　reconstruction　of　the　image　will　amplify　these　noises,　and　denoising　first　will　blur　the　image　to　a　certain　extent.　To　address　this　issue,　we　propose　a　denoising-supervised　Generative　Adversarial　Network　for　super-resolution　reconstruction　of　magnetic　resonance　images.　The　generator　of　the　network　uses　the　residual　information　distillation　module　to　separate　the　features　so　that　the　discriminator　can　guide　it　to　restore　the　image.　The　network　contains　two　discriminators,　a　sample　discriminator　and　a　noise　discriminator.　The　sample　discriminator　assists　the　generator　to　fit　the　real　sample　distribution,　and　the　noise　discriminator　assists　the　generator　to　suppress　noise.　Qualitative　and　quantitative　experiments　show　that　the　noise　discriminator　can　effectively　suppress　the　generation　of　noise　in　the　super-resolution　task.　Compared　with　advanced　methods　on　the　IXI　dataset,　the　proposed　network　achieves　the　highest　reconstruction　quality　(PSNR　and　SSIM).　©　2023,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.