A new method for generating the low-energy structures of a chain molecule was proposed recently by us. This is a stochastic process where at each step an energy-minimized structure is changed by carrying out several local torsional deformations (LTDs) along the chain, which temporarily disrupt neighbors of the rotated bonds. The energy is then minimized and the disrupted bonds return to their usual geometry (in terms of bond lengths and angles) while the chain assumes a new conformation. This conformation is accepted (and then deformed) or rejected with the help of a "selection procedure" that gives preference to accepting the lower energy structures and, thug, directs the search toward the lowest energy regions, which include the global energy minimum (GEM) structure. The selection procedures tested are the Mont Carlo minimization (MCM) method of Li and Scheraga and the "usage directed" (UD) method of Still’s group. LTD is a general method whose parameters can be optimized for any chain system. However, because of the local character of the conformational change, it is expected to be especially efficient for cyclic peptides, loops in proteins, and dense multichain systems. In this paper, LTD is applied to cycloheptadecane modeled by the MM2 force field, its parameters are optimized, and it is found to be more efficient than other methods. The results for this molecule and for an ECEPP model of the linear pentapeptide Leu-enkephalin show that MCM and UD are almost comparable in efficiency, with a slight advantage for MCM.