The field of complex self-assembly is moving toward the design of multiparticle structures consisting of thousands of distinct building blocks. To exploit the potential benefits of structures with such "addressable complexity," we need to understand the factors that optimize the yield and the kinetics of self-assembly. Here we use a simple theoretical method to explain the key features responsible for the unexpected success of DNA-brick experiments, which are currently the only demonstration of reliable self-assembly with such a large number of components. Simulations confirm that our theory accurately predicts the narrow temperature window in which error-free assembly can occur. Even more strikingly, our theory predicts that correct assembly of the complete structure may require a time-dependent experimental protocol. Furthermore, we predict that low coordination numbers result in nonclassical nucleation behavior, which we find to be essential for achieving optimal nucleation kinetics under mild growth conditions. We also show that, rather surprisingly, the use of heterogeneous bond energies improves the nucleation kinetics and in fact appears to be necessary for assembling certain intricate 3D structures. This observation makes it possible to sculpt nucleation pathways by tuning the distribution of interaction strengths. These insights not only suggest how to improve the design of structures based on DNA bricks, but also point the way toward the creation of a much wider class of chemical or colloidal structures with addressable complexity.self-assembly | free-energy landscapes | nucleation | DNA nanotechnology R ecent experiments with short pieces of single-stranded DNA (1, 2) have shown that it is possible to assemble well-defined molecular superstructures from a single solution with more than merely a handful of distinct building blocks. These experiments use complementary DNA sequences to encode an addressable structure (3) in which each distinct single-stranded "brick" belongs in a specific location within the target assembly. A remarkable feature of these experiments is that even without careful control of the subunit stoichiometry or optimization of the DNA sequences, a large number of 2-and 3D designed structures with thousands of subunits assemble reliably (1, 2, 4, 5). The success of this approach is astounding given the many ways in which the assembly of an addressable structure could potentially go wrong (6-8).Any attempt to optimize the assembly yield or to create even more complex structures should be based on a better understanding of the mechanism by which DNA bricks manage to self-assemble robustly. The existence of a sizable nucleation barrier, as originally proposed in refs. 1, 2, would remedy two possible sources of error that were previously thought to limit the successful assembly of multicomponent nanostructures: the depletion of free monomers and the uncontrolled aggregation of partially formed structures. Slowing the rate of nucleation would suppress competition among multiple nucleatio...