The invention discloses a layered self-adaptive threshold video denoising method. The method comprises the following steps: inputting a video with noise, and performing Surfacelet transformation on the video; respectively performing noise variance estimating on a coefficient decomposed by each Surfacelet layer in a direction sub-band; utilizing the size of the sub-band coefficient of the direction of each layer to calculate an initial threshold; for a high-layer decomposition sub-band, utilizing coefficient field information to perform self-adjustment on the obtained initial threshold of the layer; for other layers, utilizing a direction-energy ratio to perform self-adjustment on the initial threshold of each layer; and utilizing a soft threshold function to perform denoising processing, and reconstructing a denoised coefficient to obtain a denoised video. The method provided by the invention remarkably reduces calculating complexity, improves the PSNR value of the denoised video, can effectively keep the detail information of the video, and can be applied to natural video denoising and three-dimensional image denoising.