Methods for describing a discrete number of conformational states of amino acid residues in proteins are presented and used to investigate the topography of chain folding. The relative importance of short‐range, medium‐range and long‐range interactions is discussed in the light of an analysis of the conformational states for the different amino acid residues in eight proteins of known structure. A prediction algorithm, which assigns four states to each residue of a protein chain (α‐helix, extended structure, bend, or coil), has been developed from a consideration of both short‐ and medium‐range interactions and applied to thirteen proteins of known three‐dimensional structure. The prediction algorithm is simple to apply, and the assignment of α‐helix and extended structure is considerably better than in most other predictive schemes. The prediction of chain reversal or bend regions was also better than with previous algorithms, but these assignments were not as good as those for α‐helix and extended structure. The motivation for the development of this algorithm is not only to demonstrate the relative importance of short‐ and longer‐range interactions but, more important, to begin to develop procedures for obtaining an approximate starting conformation for subsequent energy minimization to predict the three‐dimensional structure of a protein. This procedure, as well as various other methods for the prediction of the backbone topography and conformational states of residues in proteins from the amino acid sequence, have been reviewed and evaluated by comparing the success of the methods to the success expected from a random assignment of conformational states.