Membrane-based separations have the potential to make enormous contributions in energy-related applications involving, for example, carbon dioxide capture and hydrogen purification. The fabrication and testing of non-polymeric membranes from new materials is extremely time consuming, so the risk associated with the experimental development of a genuinely new material is high. If quantitative models for the performance of new membrane materials can be developed in the absence of direct experimental data, these models can play an important role in accelerating practical development of the most promising materials. Several examples where atomically-detailed simulations have been used for this purpose will be discussed, including dense amorphous metal films for hydrogen purification and metal-organic framework materials for carbon dioxide separations.