In this work, a new methodology for automatic three dimensional shape reconstructions of bones from tomographic cross-sections is presented, based in the topological active net model [1]. Two steps fundamentally compose the methodology presented. In the first one, with the purpose of differentiate the internal and the external contours of bony structures; the existent bones in the first image of the sequence are extracted. For this purpose, a topological analysis of the mesh is done. In the second step the 3D bones' reconstruction is obtained by tracking the active net in the rest of slices using the information obtained in previous stages of the tracking process and introducing new energy terms based in directional filters.