A benchmark for non-covalent interactions in solids (C21) based on the experimental sublimation enthalpies and geometries of 21 molecular crystals is presented. Thermal and zero-point effects are carefully accounted for and reference lattice energies and thermal pressures are provided, which allow dispersion-corrected density functionals to be assessed in a straightforward way. Other thermal corrections to the sublimation enthalpy (the 2RT term) are reexamined. We compare the recently implemented exchange-hole dipole moment (XDM) model with other approaches in the literature to find that XDM roughly doubles the accuracy of DFT-D2 and non-local functionals in computed lattice energies (4.8 kJ/mol mean absolute error) while, at the same time, predicting cell geometries within less than 2% of the experimental result on average. The XDM model of dispersion interactions is confirmed as a very promising approach in solid-state applications.