Molecular dynamics is widely used to accurately sample ensembles of molecular conformations in protein folding, protein-ligand docking, and other biomolecular studies. In such simulations, it is often the case that only a region of configuration space is of interest, for example for estimating the free energy difference between conformational substates or computing a conditional average of some observable. This talk discusses a technique that we have implemented in the AMBER7 molecular dynamics code for restricting contacts through the use of reflective hypersurfaces. This technique is an extension of one originally introduced by Voter and others for use in transition state theory as applied to problems in solid state physics. Our work extends this technique to biomolecular problems and introduces a new method for performing reflections and avoiding conformational traps by the use of velocity randomization. The talk will introduce the motivation and theory for the use of reflective hypersurfaces in conformational studies, discuss two algorithms that we have implemented in AMBER7, and present molecular dynamics results using this technique to sample conditional free energies for making contacts for a small test peptide and the 56 residue IgG-binding domain B1 of protein G.