The Protein Structure Prediction problem is currently one of the most challenging open problems in Bioinformatics being a NP-Complete problem. In this work, a Multistage Simulated Annealing (MSA) employing different levels of detail for the potential energy function is applied using the Rosetta framework. The backbone and centroid coordinates model is employed being the side chains repacked at the end of the process. Experiments were conducted using four wellknown proteins with different degrees of complexity, namely: 1ZDD; 1CRN; 1ENH; 1AIL. The results obtained showed that MSA is able to find better energy function values in all four proteins, and better RMSD in three of them.