Strategies for computational association of molecular components entail a compromise between configurational exploration and accurate evaluation. Following the work of Meng et al. [Proteins, 17 (1993) 266], we investigate issues related to sampling and optimization in molecular docking within the context of the DOCK program. An extensive analysis of diverse sampling conditions for six recepto>ligand complexes has enabled us to evaluate the tractability and utility of on-the-fly force-field score minimization, as well as the method for configurational exploration. We find that the sampling scheme in DOCK is extremely robust in its ability to produce configurations near to those experimentally observed. Furthermore, despite the heavy resource demands of refinement, the incorporation of a rigidbody, grid-based simplex minimizer directly into the docking process results in a docking strategy that is more efficient at retrieving experimentally observed configurations than docking in the absence of optimization. We investigate the capacity for further performance enhancement by implementing a degeneracy checking protocol aimed at circumventing redundant optimizations of geometrically similar orientations. Finally, we present methods that assist in the selection of sampling levels appropriate to desired result quality and available computational resources.