One of the most puzzling observations in protein crystallography is that the various space-group symmetries occur with striking non-uniformity. Molecular close-packing has been invoked to explain similar observations for crystals of small organic compounds, but does not appear to be the dominant factor for proteins. Instead, we find that the observed frequencies for both two- and three-dimensional crystals can be explained by an entropic model. Under a requirement for connectivity, the favoured space groups are simply less restrictive than others in that they allow the molecules more rigid-body degrees of freedom and can therefore be realized in a greater number of ways. This result underscores the importance of the nucleation event in crystallization and leads to specific ideas for crystallizing water-soluble and membrane proteins.