Using only data on sequence, a method of computing a low-resolution tertiary structure of a protein is described. The steps are: (a) Estimate the distances of individual residues from the centroid of the molecule, using data on hydrophobicity and additional geometrical constraints. (b) Using these distances, construct a two-valued matrix whose elements, the distances between residues, are greater or less than R, the radius of the molecule. (c) Optimize to obtain a three-dimensional structure. This procedure requires modest computing facilities and is applicable to proteins with 164 residues and presumably more. It produces structures with r (correlation between inter-residue distances in the computed and native structures) between 0.5 and 0.7. Furthermore, correct inference of two or three long-range contacts suffices to yield structures with r values of 0.8-0.9. Because segments forming parallel or antiparallel folding structures intersect the radius vector at similar angles, from centroidal point distances it is possible to infer some of these long-range contacts by an elaboration of the procedure used to construct the input matrix. A criterion is also described which can be used to determine the quality of a proposed input matrix even when the native structure is not known.