Synthetic Aperture Radar (SAR) is a coherent imaging system.The images generated from SAR are often polluted by speckle noisewhich results in low accuracy in subsequent segmentation and recognition of SAR images.Aming at the problem of image pollutiona SAR image denoising network model Re-GAN is designedwhich combines Generative Adversarial Network (GAN) with the Residual Network (ResNet).The residual block of ResNet is added into the generator of GAN to enhance the ability of SAR image denoising.The combined loss function in the model can better preserve image details during noise reduction.In MATAR data setRe-GAN is compared with BM3D and wavelet denoising algorithm respectively.The experimental results show that Re-GAN has good performance in both visual effect and quantitative analysis.