Empirical potentials for interaction of proteins with intracellular ions are presented. We derive the potentials using a training dataset of the protein 3D structure bank, PDB, based on the statistical analysis of contacts between ions and protein atoms of different types. The potentials are derived using Monte Carlo Reference State, simulating non-interacting structure elements as random 3D points in the structure space. The resulting potentials are detailed, continuous, and cover a wide range of contact distances. The obtained potentials were tested for prediction of ion-binding sites in proteins and are shown to reproduce locations and specificities of ion-binding sites with a high accuracy. A web server is created for predictions of ion-binding sites in proteins.