Grand canonical Monte Carlo (GCMC) computer simulations are employed to predict selective adsorption from binary mixtures into slit, cylindrical, and spherical nanopores. The mixtures are Lennard-Jones fluids representative of Ar, Xe, and tetramethylsilane (TMS), allowing us to gauge the effects of adsorbate size, energy well depth, and mass. The adsorption selectivity can oscillate from favoring one component to the other as nanopore size increases. Beyond simple molecular sieving, two distinct mechanisms can be responsible for size selectivity. At low pore densities, selectivity depends mainly on which species is more energetically favorable in the nanopore. At high pore densities, though, the ability to pack well within the pore is most important so the component size becomes the main criterion for selectivity. In high-pore density processes (many separations, catalytic processes, and templating during zeolite synthesis), the last mechanism can be active and requires consideration.