Elastic constants of a set of molecular organic crystals have been calculated within the crystal modeling program DMAREL, which was developed to allow the use of highly accurate, anisotropic atom-atom potentials. A set of six molecular crystals (durene, m-dinitrobenzene, the form of resorcinol, pentaerythritol, urea, and hexamethylenetetramine) has been chosen to sample a range of intermolecular interactions and crystal symmetries. The sensitivity of such calculations to variations in empirical repulsion-dispersion parameters and the electrostatic model is determined and discussed relative to the other errors in the theoretical model and typical experimental uncertainties. We find that model potentials whose functional form has been simplified often reproduce crystal structures and lattice energies adequately but perform poorly when used to calculate elastic constants. This is because more theoretically justified potentials are required to satisfactorily model the curvature at the equilibrium geometries. The rigid-molecule approximation can result in exaggerated elastic constants, and the neglect of thermal effects also leads to significant overestimation of the stiffness constants.