In spite of the very large number of experiments (LEED, AES, UPS, MeV He+ scattering, work function, FIM) carried out on W (100), no structural model consistent with all the data has been proposed yet: in particular, the question of the reconstruction thermally induced when the sample is cooled below room temperature remains a puzzling problem. Furthermore, from a theoretical point of view, no definitive answer has been given. Actually, either the mechanism invoked for the reconstruction is too weak, or some contributions are omitted or calculated without sufficient care. The authors compute here the surface energy of W (100) taking into account the band term treated in the tight binding approximation, a pairwise repulsive potential of the Born-Mayer type and the electronic correlation contribution obtained from a perturbation treatment of the Hubbard model in the band limit. This energy is then fully minimised with respect to all coordinates of surface atoms, keeping all atoms neutral for any displacement. They find that the unreconstructed surface is the most stable at T=0K and discuss this unexpected result. A similar calculation for Ta (100) leads to opposite conclusions.