Reliable high-resolution prediction of protein complex structures starting from the free monomers is a considerable challenge toward large-scale mapping of the structural details of protein-protein interactions. The current major bottleneck is to model the conformational changes of the monomer backbone upon binding. We evaluate the use of homolog structures as source for conformational diversity, within the framework of RosettaDock-a leading high-resolution docking protocol. We find that the use of homolog templates can improve significantly the modeling of a complex structure, including known difficult cases. Several conformational changes however are not sampled by any of the templates, indicating the need for additional sources of conformational variability. Interestingly, the successful homolog templates are not restricted to a confined range of sequence identity, highlighting the importance of the backbone conformation rather than the sequence.