This paper proposes a new, fast, triangular mesh approximation method for the 3D visualization of surface data. Using spatio-frequency localization characteristics and directional information of wavelet coefficients, we determine local complexities of surface data and approximate the data to a proper triangular mesh. The proposed algorithm is quite simple, and the computational cost is low due to the direct use of wavelet coefficients for vertex removal. The computer simulation results for terrain data show that the proposed algorithm is excellent for fast 3D visualization.