Pore-resolved computations are undertaken, within a continuum model framework, to explore surface diffusion as a selective mechanism for gas separations using membranes of randomly overlapping parallel cylindrical pores or fibers. Orders of magnitude of the model parameters are established using an intrinsic gas diffusivity that is self consistent with the Knudsen diffusivity obtained from Monte Carlo simulations reported in the literature. The relative contributions of surface and gas diffusion to the overall permeation show that the surface-diffusion flux increases with the specific surface area, whereas the gas flux increases with porosity. Thus, gas diffusion that is perpendicular to pores and fibers can be hindered by the increasing tortuosity while simultaneously promoting, permeation via surface diffusion. The selectivity of pore structures with fibrous networks is examined for the dehumidification of air, natural gas, and carbon dioxide. Selectivities, defined as the ratio of the effective diffusion coefficients for the adsorbed phases (moisture) and the, void-diffusing gas (air, natural gas, or carbon dioxide), reveal' that selectivity is higher for (i) heavier gases, which have lower gas-diffusion fluxes, and (ii) capillary pores, which have higher surface connectivity than fibers.