Recently, high free volume polymer materials have been regarded as high potential candidates for gas transport/separation membranes, since the amount of free volume in polymeric membrane can improve the diffusivity and solubility of gas molecules. In this study, we focused on how local changes in polymer structure can affect the performance of a membrane at the molecular level. The transport behavior was theoretically analyzed, and then the differences in the amount and morphology of free volume were characterized. Finally, we suggested how the "evolution of microcavities" affects the gas transport properties of hydroxyl-containing polyimide (HPI) and thermally rearranged (TR) polymers. In particular, using image analysis, we intuitively demonstrate the morphological difference between HPI and TR polymers that have been indirectly explained by experimental analyses using a wide-angle X-ray diffractometer (WAXD) and positron annihilation laser spectroscopy (PALS). Solubility results using the grand canonical Monte Carlo (GCMC) method showed marginal improvement in thermally rearranged polybenzoxazoles (TR-PBOs) from its precursor HPI, which is in good agreement with the experimental tendency. Moreover, higher diffusivities but lower selectivities of TR-PBO models compared with those of HPI models were observed, as reported experimentally. The difference in gas transport abilities between HPIs and TR-PBOs originates from the difference in their diffusion behavior, and this is strongly related to the free volume amount and morphology of polymeric materials. In addition to the higher amount of total free volume in TR-PBO, our image analysis revealed that TR-PBO has a higher amount of interconnected "hourglass-shaped free volume elements", which consist of larger and more elongated cavities with bottlenecks than the HPI model. In particular, the bottleneck diameters in the TR-PBO models are wider than those in the HPI models, enabling the larger gas molecules to diffuse through the cavities faster. However, the narrower and smaller bottleneck diameters in the HPI model can induce better selectivity for large gas molecules.