Conformational changes in proteins are intrinsic to almost every function, but sampling such changes remains one of the most difficult bio-computational problems. The challenge lies on two levels. First, algorithms which can efficiently focus computer time on the correct path ensemble are necessary. Second, even if perfect path-sampling algorithms were available, intrinsic "fast" timescales of protein motions are too costly for atomistic models; coarse-grained models are therefore required in most systems of interest. The group's work on both these issues will be described. Algorithmically, the "weighted ensemble" method is generalized and set on a firm statistical footing. Toward the goal of determining the most chemically realistic models which can be fully sampled, we describe a novel class of semi-atomistic models, based on pre-calculated residue libraries.