The natural amino acids have different preferences of occurring in specific types of secondary protein structure. Simulations are performed on periodic model beta-sheets of 14 different amino acids, at the level of density functional theory, employing the generalized gradient approximation. We find that the statistically observed beta-sheet propensities correlate very well with the calculated binding energies. Analysis of the calculations shows that the beta-sheet propensities are determined by the local flexibility of the individual polypeptide strands.