An approach to ab initio crystal structure prediction by packing optimization is developed for organic nitramines, an important class of energetic materials. The principal features of the search method are: use of statistical data on the organic crystal structural classes to select typical space groups and site symmetries for further search; accounting for the energy-hypersurface symmetry to determine the unique search region; and use of an automated similarity-search procedure to recognize non-unique minima and determine the symmetry of optimized packings. The wide convergence properties of the local search procedure permit one to start optimization from an arbitrary point, so that no preliminary screening of the starting models is necessary. The numerical calculations were first carried out on the known crystal structures of three polymorphs of HMX (1,3,5,7-tetranitro-1,3,5,7-tetraazacyclooctane). In this step, the force-field parameters for the nitramine fragment have been improved to obtain the best correspondence between the predicted and observed molecular geometries. The predicted packings were found to be in reasonably good agreement with the X-ray structural data, while the computed lattice energies were not accurate enough to predict the observed heats of sublimation and the trend of polymorph stabilities. Secondly, the method was employed to predict the possible crystal structures of eight isomeric azanitroadamantanes and wurtzitanes, whose molecular structures were proposed earlier on the basis of a computational study (T.S. Pivina et al., Propellants, Explosives, Pyrotechnics, 20 (1995) 91). As a result, the energy-minimized structures with densities up to 2.08 and 2.04 g cm - 3 have been predicted for the adamantane and wurtzitane series, respectively, as the possible crystal polymorphs. Due to the interaction between the conformational and packing forces giving rise to some gain in the total energy at the expense of at least partial loss in molecular symmetry, the predicted densities are expected to be lower estimates than the actual ones.