A new Monte Carlo algorithm is presented for the efficient sampling of protein conformation space called the Smart-Walking (S-Walking) method. The method is implemented using a hybrid Monte Carlo protocol. The S-Walking method is closely related to the J-Walking method proposed by Frantz et al. (J. Chem. Phys. 93, 2769, 1990). Like the J-Walking method, the S-Walking method runs two walkers, one at the temperature of interest, the other at a higher temperature which more efficiently generates ergodic distributions. Instead of sampling from the Boltzmann distribution of the higher temperature walker as in J-Walking, S-Walking first approximately minimizes the structures being jumped into, and then uses the relaxed structures as the trial moves at the low temperature. By jumping into a relaxed structure, or a local minimum, the jump acceptance ratio increases dramatically, which makes the protein system easily undergo barrier-crossing events from one basin to another, thus greatly improving the ergodicity of the sampling. The method approximately preserves detailed balance provided the time between jumps is large enough to allow effective sampling of conformations in each local basin. (C) 1997 American Institute of Physics.