The invention discloses a Bayesian denoising method based on wavelet low frequency. The method comprises the following steps of: 1, inputting a natural image to be denoised; 2, selecting pixel blocks to be estimated; 3, selecting a central low-frequency-coefficient block; 4, determining a searching window; 5, selecting a low-frequency-coefficient block; 6, judging whether a constraint condition is satisfied; 7, computing a similarity weight value; 8, judging whether all points in the searching window are searched; 9, computing the recovery values of the pixel blocks to be estimated; 10, judging whether the natural image to be denoised is searched completely; and 11, integrating the recovery values. The similarity weight value is computed by using a wavelet low frequency coefficient. Compared with the conventional denoising method, the Bayesian denoising method has the advantages that: the edge and texture details of the natural image can be well kept and recovered while noise is well smoothened; and the Bayesian denoising method can be applied to denoising processing of the natural image.