This thesis describes an algorithm that takes height field data as an input, and efficiently approximates it with irregular triangle meshes. The basic idea is to use the wavelet coefficient as an importance measure of the corresponding surface region. It starts with a coarse approximation and adds more faces to regions having large wavelet coefficients to reduce the error. For the ease of implementaion, it uses three simple transformations of "mesh optimization". The algorighm is efficient due to its top-down approach, and provides good performance by the localization property of wavelets.