Local surface reconstruction by the Marching Cubes algorithm and its derivatives has a well known ambiguity, which prevents constructed surfaces from being closed and simple. We investigate this ambiguity assuming that a 3D image samples well‐defined objects. In this case it is justified to aim at tiling of extracted object voxels rather than at reconstructing iso surfaces. Compared to iso surface reconstruction, our algorithm provides essentially the same level of confidence with respect to surface location at a lower computational cost. We present a leak detection and mending scheme which resolves the Marching Cubes ambiguity and guarantees a well‐defined behaviour with respect to which objects are covered by which surface. We detail how to implement our leak mending method within a completely tabulated Marching Cubes algorithm. We finally give an example of how the adapted algorithm is of benefit to a recently developed 3D MR spectroscopy technique.