Abstract : The theory of the relativistic mean-field (RMF) has been considered for several years as one of the most promising for describing the atomic nuclei. However, many approximations are made in this approach, and several of them are not kept under control. For this reason, we preferred to develop a slightly different formalism, based on a parametrisation of the nuclear potentials, which exhibits numerous advantages (simplicity, reliability, connection with the group theory of spinors, etc.). This approach enabled us to prove for the first time the non-uniqueness of the spin-orbit interaction mechanism, as well as the influence of the effective mass on the one-body properties of atomic nuclei. Since our goal was also to build a hamiltonian that would be really efficient for all the applications encountered in nuclear structure, the parameters of the potentials have been fitted very thoroughly to the most recent experimental data according to a multi-dimensional minimization procedure. The results show an excellent stability as functions of isospin and nuclear mass. In spherical nuclei, the positions of the individual levels are significantly better reproduced when using the new approach - as compared to the other theories available today. It turns out that the relativistic theory is essential for understanding the effective inertia of deformed atomic nuclei. Several other applications have beeen explored : among them, tetrahedral symmetry is predicted to be a systematic feature of the shell structure in nuclei, and is therefore present from the lightest up to the super-heavy elements. Our first calculations suggest that the latter could preferably stabilize in such a tetrahedral shape rather than in a spherical one.