Background Although the accumulation of protein network data in a wide range of species provides a rich resource for understanding network evolution, exploiting such resources remains challenging. Difficulty in defining orthologous relationships and the noisy and incomplete nature of protein network data  are both major obstacles. To overcome the first, we took a domain-based view of the proteome, with domains of known (3D) structure as nodes in the network. This not only provides a structural basis for the interactions, but simplifies the question of orthology. The second is mitigated by an FDR-based statistical inference of the underlying protein domain network from multiple domain interactions between all proteins in the genome. Thus the negative influence of noisy interactions is reduced. The comparisons of these domain networks are also less susceptible to low coverage of networks, especially in lessstudied species. Materials and methods Accordingly, here we present a statistical inference of domain networks from multiple sources of protein interaction data taken from STRING , combined with domain compositions of proteins from SUPERFAMILY  across hundreds of species. This is followed by an intersection analysis for comparing domain networks between any two species using a third as reference (thus accounting for differences in coverage).