Delaunay tessellation followed by Monte Carlo integration is employed in order to determine the clusters of sites where a hard-sphere penetrant of radius r p equal to a few Angstroms can reside in model carbon nanotube-atactic poly-(methyl-methacrylate) (CNT-PMMA) nanocomposite microstructures and analyze their dependence on penetrant size and temperature. Starting configurations for the geometric analysis are generated by cooling down to lower temperatures the atomistic structures fully equilibrated at a higher temperature by means of a long molecular dynamics simulation and re-equilibrating. Because the tetrahedra formed in the process of the Delaunay tessellation are irregular in space, an analytical calculation of free volume is a tough problem; to overcome this, we resort to Monte Carlo integration. By accounting for the volume occupied by polymers and CNT atoms, we obtain estimates of the unoccupied volume as well as of the volume accessible to a spherical penetrant of a given radius within each tetrahedron. From this, we calculate next the distribution of the volume and size of the corresponding cavities and of their clusters. By identifying neighboring clusters of tetrahedra that are mutually accessible to a given penetrant using a connectivity algorithm very similar to that proposed by Greenfield and Theodorou [Macromolecules, 1993, 26, 5461−5472], we quantify the network of clusters formed and determine probable pathways for diffusion for the penetrant under study.