Surfaces reconstruction from serial section images is a mature research in visualization of medical imaging. Traditional methods always adopt the process that starts on mesh reconstruction or isosurfaces extraction, then data simplification. Since the volumetric datasets are huge and segmented. Most of the existing algorithms that use large in-core data structures might be too large to off-load to disk. In this paper, an algorithm is presented that has been designed to reconstruct 3D surfaces from huge and segmented volumetric datasets. Approximate mesh can be reconstructed and simplified layer by layer. And user can control mesh complexity by parameters. Because memories are used rationally, some surfaces reconstruction from large datasets can work on PC using the algorithm.