Crystallization is a fundamental and ubiquitous process much studied over the centuries. But although the crystallization of atoms is fairly well understood, it remains challenging to predict reliably the outcome of molecular crystallization processes that are complicated by various molecular interactions and solvent involvement. This difficulty also applies to nanoparticles: high-quality three-dimensional crystals are mostly produced using drying and sedimentation techniques that are often impossible to rationalize and control to give a desired crystal symmetry, lattice spacing and habit (crystal shape). In principle, DNA-mediated assembly of nanoparticles offers an ideal opportunity for studying nanoparticle crystallization: a well-defined set of rules have been developed to target desired lattice symmetries and lattice constants, and the occurrence of features such as grain boundaries and twinning in DNA superlattices and traditional crystals comprised of molecular or atomic building blocks suggests that similar principles govern their crystallization. But the presence of charged biomolecules, interparticle spacings of tens of nanometres, and the realization so far of only polycrystalline DNA-interconnected nanoparticle superlattices, all suggest that DNA-guided crystallization may differ from traditional crystal growth. Here we show that very slow cooling, over several days, of solutions of complementary-DNA-modified nanoparticles through the melting temperature of the system gives the thermodynamic product with a specific and uniform crystal habit. We find that our nanoparticle assemblies have the Wulff equilibrium crystal structure that is predicted from theoretical considerations and molecular dynamics simulations, thus establishing that DNA hybridization can direct nanoparticle assembly along a pathway that mimics atomic crystallization.