The conformations of proteins and protein−protein complexes observed in nature must be low in free energy relative to alternative (not observed) conformations, and it is plausible (but not absolutely necessary) that the electrostatic free energies of experimentally observed conformations are also low relative to other conformations. Starting from this assumption, we evaluate alternative models of electrostatic interactions in proteins by comparing the electrostatic free energies of native, nativelike, and non-native structures. We observe that the total electrostatic free energy computed using the Poisson−Boltzmann (PB) equation or the generalized Born (GB) model exhibits free energy gaps that are comparable to, or smaller than, the free energy gaps resulting from Coulomb interactions alone. Detailed characterization of the contributions of different atom types to the total electrostatic free energy showed that, although for most atoms unfavorable solvation energies associated with atom burial are more than compensated by attractive Coulomb interactions, Coulomb interactions do not become more favorable with burial for certain backbone atom types, suggesting inaccuracies in the treatment of backbone electrostatics. Sizable free energy gaps are obtained using simple distance-dependent dielectric models, suggesting their usefulness in approximating the attenuation of long range Coulomb interactions by induced polarization effects. Hydrogen bonding interactions appear to be better modeled with an explicitly orientation-dependent hydrogen bonding potential than with any of the purely electrostatic models of hydrogen bonds, as there are larger free energy gaps with the former. Finally, a combined electrostatics−hydrogen bonding potential is developed that appears to better capture the free energy differences between native, nativelike, and non-native proteins and protein−protein complexes than electrostatic or hydrogen bonding models alone.