The superatom model for nanoparticle structure is shown to be inadequate for the prediction of the thermodynamic stability of gold nanoparticles. The observed large HOMO-LUMO gaps for stable nanoparticles predicted by this model are, for sulfur-stabilized gold nanoparticles, attributed to covalent interactions of the metal with thiyl adsorbate radicals rather than ionic interactions with thiolate adsorbate ions, as is commonly presumed. In particular, gold adatoms in the stabilizing layer are shown to be of Au(0) nature, subtle but significantly different from the atoms of the gold core owing to the variations in the proportion of gold-gold and gold-sulfur links that form. These interactions explain the success of the superatom model in describing the electronic structure of both known and informatory nanoparticle compositions. Nanoparticle reaction energies are, however, found not to correlate with the completion of superatom shells. Instead, local structural effects are found to dominate the chemistry and in particular the significantly different chemical properties of gold nanoparticle and bulk surfaces. These conclusions are drawn from density-functional-theory calculations for the Au(102)(p-mercaptobenzoic acid)(44) nanoparticle based on the X-ray structure (Jadzinsky, P. D.; et al. Science 2007, 318, 430), as well calculations for the related Au(102)(S(*)-CH(3))(44) nanoparticle, for the inner gold-cluster cores, for partially and overly reacted cores, and for Au(111) surface adsorbates.