Recently developed atomistic models of highly disordered nanoporous materials offer hope for a much more realistic description of the pore morphology and topology in such materials; however, a factor limiting their application has been the computationally intensive characterization of the models, particularly determination of the pore size distribution. We report a new technique for fast computation of pore size distributions of model materials from knowledge of the molecular coordinates. The pore size distribution (PSD) is defined as the statistical distribution of the radius of the largest sphere that can be fitted inside a pore at a given point. Using constrained nonlinear optimization, we calculate the maximum radii of test particles at random points inside the pore cavity. The final pore size distribution is then obtained by sampling the test particle radii using Monte Carlo integration. The computation time depends on factors such as the number of atoms, the sampling resolution, and the desired accuracy. However, even for large systems, PSDs with very high accuracy (>99.9%) are obtained in less than 24 h on a 3 GHz Pentium IV processor. The technique is validated by applying it to model structures, whose pore size distributions are already known. We then apply this method to investigate the pore structures of several mesoporous silica models such as SBA-15 and mesostructured cellular foams.