Abstract: In order to avoid the ambiguity of isosurface extraction process in 3D mesh, an isosurface extraction algorithm based on gradient in 3D data space was proposed. In this method, contour points were calculated on the surface of the voxel based on gradient step by step. The outline of the isosurface was drew to make sure that the isosurface was concave or convex by adding contour points inside the voxel. The map-Delaunay method was proposed to get triangles of the isosurface in 2D. The step and merge-threshold were used to control the quantity of triangles to decrease the cost of storage. The experiment results showed that with this algorithm, the accurate contour was extracted just for once and the shortcomings of Marching Cubes algorithm was overcome. The isosurface in the voxel could be extracted exactly. The accuracy of the isosurface was ensured, and the cost of storage was reduced by changing the parameters.