A permeability - selectivity trade-off is observed in kinetic (diffusion-based) separations, such as oxygen/nitrogen, on polymeric and inorganic microporous membrane materials. Some materials are inherently better than others in the trade-off, but the molecular-level reasons are not clear. In this paper, we employed a molecular modeling approach to study the oxygen/nitrogen separation and provide insight into optimizing the pore characteristics (geometry, material of construction) for separation performance. The pore models were abstract, with features representative of different microporous or polymeric materials. They comprised three-dimensional atomistic models of unidirectional pores with diameters <5 Å; various aspects of the pore geometry and material of construction were varied. Oxygen and nitrogen were modeled as rigid dimers, and the Lennard-Jones (LJ) potential was used to model the gas - pore site interactions. Free energy calculations and transition-state theory were used to predict the permeation rates through the pores. Better performance on the permeability - selectivity plot is obtained for pores (1) constructed of atoms with deeper LJ attractive wells, (2) having a variable cross section (wider regions joined by narrower necks), and (3) having a higher number of atoms in the pore perimeter. The trade-off curve was also sensitive to the choice of LJ potential parameter set.