This paper describes a new method for searching the potential energy hypersurface of organic crystals. Clusters made of 2-4 molecules are built with the most common symmetry elements: the inversion center, the screw axis, and the glide plane. The relative importance of these elements, as well as of lattice translation, in building the cohesive energy of the crystal, is assessed by using a statistical analysis of known hydrocarbon structures. A full-crystal structure is then built by translation of these clusters, yielding the most frequent space groups for organic compounds. Cluster and lattice energies are calculated by using empirical potentials; use is also made of statistical correlations between molecular and crystal properties to guide the search. Tests on known crystal structures of hydrocarbons show that in some favorable cases the method can predict the crystal structure from molecular structure only; for one compound, a prediction of the unknown crystal structure is given. The results demonstrate quantitatively that many crystal structures with small energy differences are possible for moderately polar organic molecules.