We propose a new and efficient denoising method for remote sensing image based on the contourlet transform. Firstly, we introduce a geometrical prior model, and estimate the Bayesian framework by combining the noise with the useful signal. Then, we compute each coefficient as the probability of evailable signal, modifying the wavelet shrinkage factor. Finally, we reconstruct multi-resolution wavelet coefficient and process it using recursive cycle spinning. The experimental results show that the denoised image eliminates pseudo-Gibbs phenomena, the peak signal-to-noise ratio increases 1～2 dB.