The problem of protein folding has become the subject of intense theoretical and experimental study over the past few decades. This work presents a new method for protein structure prediction using distance and dihedral angle restraints derived from NMR data. Both the formulation and solution approach differ substantially from traditional molecular structure prediction techniques. The traditional formulation is recast as a constrained global optimization problem whose solution is obtained through the use of the alpha BB algorithm, a deterministic global optimization approach suitable for nonconvex constrained problems. To enhance the efficiency of this method, torsion angle dynamics is introduced as an integral part of the solution approach. The proposed algorithm is tested on the Compstatin structure prediction problem.