A novel image denoising algorithm based on nonsubsampled Contourlet transform domain is proposed. First, according to the correlation of nonsubsampled Contourlet transform coefficients in interscale and intrascale, nonGaussian distribution model is used to model its correlations. We propose a classification standard where the coefficients are divided into important and unimportance coefficients, and generalized Gaussian distribution is used to describe the probability distribution for the important coefficients. Adaptive threshold is derived under the Bayesian theory and the best range of the parameter is found out. In order to overcome the shortcoming of the soft and hard thresholding function, then a new adjustable thresholding function is presented. Lastly, the new thresholding function is used to estimate coefficients without noise, and inverse nonsubsampled Contourlet transformation is performed to get denoised image. Experimental results show that our algorithm outperforms the other current outstanding algorithms in peak signaltonoise ratio, structural similarity and visual quality.