3D surface reconstruction is a key problem in computer vision. To obtain the 3D surface reconstruction of an object from its 2D image sequence one needs to get dense matching between these images. However the pixels constituting the 2D images have an unequal importance. When people watching a scene, his eyes will been firstly focused on the outlines of the object and then on its details, a fact corresponding with the characteristic of human vision. So, it is reasonable to introduce a coarse-to-fine strategy to 3D surface reconstruction. In this paper we propose a hierarchical approach. Based on Wavelet transformation of the 2D images, we can get the dense matching in different levels. As a result, we can get 3D surface reconstruction with various approximation qualities. Experiments with real images show that our algorithm is feasible.