There are two aspects of the theoretical approach to the protein folding problem. The first is to compute the thermodynamically stable native structure, and the second is to compute the folding pathways from the unfolded to the folded native form. The evolution of physics-based computational methodology from an all-atom representation of the polypeptide chain to a united-residue representation of the chain will be discussed. Blind tests in successive CASP exercises demonstrate increasing prediction success, in computing protein structure, from one CASP test to another. As for folding pathways, two different methods are used: (1) a stochastic difference equation procedure, and (2) Lagrangian dynamics with the united-residue force field. The results of all the computations, and the methods leading to them, will be discussed.