Noise estimation can be applied to image denoising, image restoration and image quality assessment and so on. The main performance of noise for remote sensing image is Gaussian noise. Due to high complexity of its texture, it is very difficult to estimate noise. For remote sensing image, a noise estimation algorithm was proposed based on layered mathematical morphology. Firstly, stable regions of different layer were extracted by close-operation with different structuring elements in binary edge image. On this basis, noise value was estimated in different layer. Secondly, the best separation layer of noise and signal was determined by analyzing the variation of noise variance of adjacent layers. Finally, noise value was estimated by specific rules based on the properties of signal-to-noise separation layer and its adjacent layer. Experimental results show that the proposed method can effectively solve the problem of block inaccuracy compared with traditional algorithms. The noise standard deviation relative error is less than traditional algorithms. It can be used in images with complex texture and has a wider range of application.