The challenge of conventional parametric model-based wavelet image denoising approaches is that the efficiency of these methods greatly depends on the accuracy of the prior distribution used for modelling the wavelet coefficients. To tackle this challenge, a non-parametric statistical model is proposed in this paper to formulate the marginal distribution of wavelet coefficients. The proposed non-parametric model differs from conventional parametric models in that the proposed model is automatically adapted to the observed image data, rather than imposing an assumption about the distribution of the data. Furthermore, the proposed non-parametric model is incorporated into a Bayesian inference framework to derive a maximum a posterior estimation based image denoising approach. Experiments are conducted to demonstrate the superior performance of the proposed approach.