In this study we report on the application of the diffusion process-controlled Monte Carlo method to a 20 amino acid alpha beta peptide (Ac-E-T-Q-A-A-L-L-A-A-Q-K-A-Y-H-P-M-T-M-T-G-Am). The polypeptide chain is represented by a set of 126 particles, the side chains are modeled by spheres, and the backbone dihedral angles phi and psi of each of the amino acid residue are essentially restricted to a set of ten high probability regions, although the whole phi-psi, space may be visited in the course of the simulation. The method differs from other off-lattice Monte Carlo methods, in that the escape time from one accepted conformation to the next is estimated and limited at each iteration. The conformations are evaluated on the basis of pairwise nonbonded side chain energies derived from statistical distributions of contacts in real proteins and a simple main chain hydrogen bonding potential. As a result of four simulations starting from random extended conformations and one starting from a structure consistent with NMR data, the lowest-energy conformation (i.e., the alpha beta fold) is detected in similar to 10(3) Monte Carlo steps, although the estimated probability of getting the alpha beta motif is similar to 10(-12). The predicted conformations deviate by 3.0 Angstrom rms from a model structure compatible with the experimental results. In this work further evidence is provided that this method is useful in determining the lowest-energy region of medium-size polypeptide chains.