We present an approach for the development of knowledge-based mean-force potentials for representing residue-residue interactions in proteins at the amino acid level of resolution which take into account the rotamer states of the residues. The rotamer-specific potentials which consist of an additional correction term to the current database-derived statistical potentials are expressed in terms of distance-dependent mean force potentials (MFP). To describe the relative geometry of two residues we used a set of six distances between different pairs of atoms. The distant-dependent MFPs are calculated from computer simulations of each pair of amino acids at the atomic level using the weighted-histogram analysis method. The umbrella sampling for each of 52003 different rotamer pairs was performed with a force-bias Monte Carlo algorithm using molecular mechanics potentials with statistically derived restraint terms. Our simulations show that the MFPs are very sensitive to the rotamer states of the amino acid residues. Preliminary tests of the potential＇s performance show that it can reproduce side-chain packing with reasonable accuracy.