We describe a novel, two-step method for directly computing the conformational free energy of a molecule. In the first step, a finite set of low-energy conformations is identified, and its contribution to the configuration integral is evaluated by a straightforward Monte Carlo technique. The method of finding energy minima incorporates certain features of the global-underestimator method and of a genetic algorithm. In the second step, the contribution to the configuration integral due to conformations not included in the initial integration is determined by Metropolis Monte Carlo sampling. Applications to alanine oligopeptides and to three cyclic urea inhibitors of HIV protease are presented.