Key to successful protein structure prediction is a potential that recognizes the native state from misfolded structures. In this thesis, we introduced a novel way to extract interaction potential functions between the 20 types of amino acids, which used the Modified Hypenetted Chain (MHNC) and the Reverse Monte-Carlo (RMC) method. We extract Radial Distribution Functions (RDFs) from 996 known protein crystal structures from the Protein Data Bank, and using these RDFs we were able to first generate the potential-of-mean-force (PMF) for different pairs of residues, and then we improved these PMFs by including the higher order terms of the Ornstein-Zernike equation using an iteration that starting from the HNC approximation for the pair interaction potential, and in each of the follow step, we conducted Monte-Carlo simulations to generate the RDFs for the updated potential. The updated potentials in each iteration step can be generated either using MHNC or the RMC method. These effective pairwise potentials were then summed up to obtain the total energy score for known protein structures, and their effectiveness was validated by conducting single and multiple decoy set tests using the `R' Us decoy set.