Several geometric-based methods have been developed for the last two to three decades to detect and identify cavities (i.e., putative binding sites) on proteins, as needed to study protein-ligand interactions and protein docking. This paper introduces a new protein cavity method, called CavVis, which combines voxelization (i.e., a grid of voxels) and an analytic formulation of Gaussian surfaces that approximates the solvent-excluded surface. This method builds upon visibility of points on protein surface to find its cavities. Specifically, the visibility criterion combines three concepts we borrow from computer graphics, the field-of-view of each surface point, voxel ray casting, and back-face culling.