Protein conformation families for automatic model building were determined for dipeptidic, tripeptidic, tetrapeptidic and pentapeptidic fragments. Mapping in n-dimensional conformational space (n = 2, 4 and 6), a conformation-generator method, a deletion-sorting process and a verification procedure were used to calculate the conformational preferences. Torsion angles were harvested from PDB structures with resolutions better than 1.5 A. Statistical weights were calculated for the conformation families.