Growing high-quality crystals is a bottleneck in the determination of protein structures by x-ray diffraction. Experiments find that materials with a disordered pitted surface seed the growth of protein crystals. Here we report computer simulations of rapid crystal nucleation in nanoscale pits. Nucleation is rapid, as the crystal forms in pits that have filled with liquid via capillary condensation. Surprisingly, we find that pits whose surfaces are rough are better than pits with crystalline surfaces; the roughness prevents the growing crystal from trying to conform to the pit surface and becoming strained. DOI: 10.1103/PhysRevLett.105.205501 PACS numbers: 81.10.Aj, 64.60.QÀ, 64.70.dg, 87.15.nt The faster crystals grow, the more defects they contain. Hence, high-quality protein crystals that can be used in xray or neutron diffraction studies are ideally grown at low supersaturation where crystal growth is slow. Yet, to nucleate a protein crystal from solution, the supersaturation cannot be too small, because then crystals never form as nucleation is effectively suppressed. The challenge for protein crystallization is therefore to nucleate crystals at supersaturations that are so low that crystal growth is slow. One possible strategy to achieve this objective is to make use of heterogeneous nucleation: a suitable nucleation agent can induce crystallization under conditions where homogeneous (bulk) crystal nucleation is negligible and crystal growth is slow. The design of suitable crystalnucleating agents is therefore of great practical importance.The rational design of such agents is hampered by a lack of understanding of the molecular mechanism of nucleation. One might think that the best strategy is to use structured templates that match the lattice spacing of the target crystal. Indeed, experiments on colloidal systems have shown that a template with a periodic surface pattern can strongly enhance crystal growth via epitaxy [1]. However, unless there is a precise match between the lattice spacing of the template and that of the epitaxially grown crystal, epitaxial crystal growth is not possible. Thus templates tend to work only for specific target crystals that are commensurate with the periodicity of the template. Such highly specific templates are of little use if we want to crystallize a wide variety of proteins with different and a priori unknown crystal lattices.An alternative to templates was pioneered by Chayen and co-workers [2-4], who showed that materials with disordered nanoscale pitted surfaces can act as ''universal'' nucleation agents in the sense that they induce the nucleation of crystals of many different proteins. See Refs. [5,6] for reviews of protein crystallization.In this Letter we present Monte Carlo simulations that investigate the mechanism of protein crystallization in nanoscale pits. Our aim is to study heterogeneous crystal nucleation in a simple model system with a phase diagram similar to that of many globular-protein solutions, such as lysozyme. Our computer simulations reve...