This article discusses the use of molecular replacement (MR) with Phaser to solve protein complex structures. MR becomes more challenging as the number of components in the asymmetric unit increases. Maximum-likelihood MR functions in Phaser enable complex asymmetric units to be built from individual components using a 'tree search with pruning' approach, which has been successful in solving many previously intractable MR problems. However, there are cases where the automated search procedure is suboptimal, such as when there are many copies of the same component or when components have greatly varying B factors. Two case studies illustrate how Phaser can be used effectively in automated MR mode, while two others show how to modify the search strategy for problematic cases. The article describes the automated MR process in Phaser, which includes anisotropy correction, model generation, rotation function, translation function, packing function, and rigid-body refinement. The 'tree search with pruning' strategy is used to search for multiple components in the asymmetric unit, with the best solution being identified by the highest signal-to-noise ratio. The success of MR depends on the fraction of the asymmetric unit with a suitable model and the r.m.s. deviation between the model and target structures. The r.m.s. deviation generally increases with decreasing sequence identity, so good models generally have high sequence identity with the target structure. The article presents four case studies: BETA–BLIP, ROP four-helix bundle, Vκ antibody fibre, and AP2 complex. These examples illustrate how Phaser can be used to solve complex structures, including cases where the model has low sequence identity or where components have different B factors. The case studies show how the 'tree search with pruning' strategy can be used to search for multiple components in the asymmetric unit, and how the search can be modified for problematic cases. The article concludes that the maximum-likelihood MR functions in Phaser have enabled many previously intractable MR problems to be solved, and that future versions of Phaser should overcome the shortcomings of the current version.This article discusses the use of molecular replacement (MR) with Phaser to solve protein complex structures. MR becomes more challenging as the number of components in the asymmetric unit increases. Maximum-likelihood MR functions in Phaser enable complex asymmetric units to be built from individual components using a 'tree search with pruning' approach, which has been successful in solving many previously intractable MR problems. However, there are cases where the automated search procedure is suboptimal, such as when there are many copies of the same component or when components have greatly varying B factors. Two case studies illustrate how Phaser can be used effectively in automated MR mode, while two others show how to modify the search strategy for problematic cases. The article describes the automated MR process in Phaser, which includes anisotropy correction, model generation, rotation function, translation function, packing function, and rigid-body refinement. The 'tree search with pruning' strategy is used to search for multiple components in the asymmetric unit, with the best solution being identified by the highest signal-to-noise ratio. The success of MR depends on the fraction of the asymmetric unit with a suitable model and the r.m.s. deviation between the model and target structures. The r.m.s. deviation generally increases with decreasing sequence identity, so good models generally have high sequence identity with the target structure. The article presents four case studies: BETA–BLIP, ROP four-helix bundle, Vκ antibody fibre, and AP2 complex. These examples illustrate how Phaser can be used to solve complex structures, including cases where the model has low sequence identity or where components have different B factors. The case studies show how the 'tree search with pruning' strategy can be used to search for multiple components in the asymmetric unit, and how the search can be modified for problematic cases. The article concludes that the maximum-likelihood MR functions in Phaser have enabled many previously intractable MR problems to be solved, and that future versions of Phaser should overcome the shortcomings of the current version.