Abstract: Curvelet is a new multiscale transform theory,which has the characteristics of anisotropy.It can approach a high dimensional function containing line singularity better.Based on curvelet decomposition,Bayesis estimation is obtained by the estimate rule derived from Bayesis theory is obtained.A new adaptive method of image de-noising based on the curvelet domain and empirical Bayesis estimation is proposed.The experiments show that compared with the other de-noising methods,the proposed approach can obtain better visual quality and improve objective measurements,which can demonstrate the advantage under the larger noise situation.