Knowledge of the three-dimensional (3D) structure of a protein-protein complex provides insights into the function of the system that can guide, for example, the systematic design of novel regulators of activity. However, at the end of 1997, there were more than 5000 protein structures in the Brookhaven databank (PDB) but less than 200 sets of coordinates for protein-protein complexes. This disparity is reminiscent of the protein-sequence/protein-structure gap and similarity motivates the development of computational methods for structure prediction. This chapter describes the strategy to start with the coordinates of the two molecules in their unbound states and then computationally model the structure of the bound complex including the conformational changes on association. For reviews of the field of protein docking see refs. 1–3.