A line of research (e.g. [1]) has considered the question of self-similarity and self-dissimilarity in complex networks.
Very generally, a self-similar object is similar to a part of itself. A natural application of this definition to networks would be to say that a network is self-similar if its graph is composed of subgraphs that are structurally similar to the graph as a whole. Still, this is not yet a precise, quantifiable definition - for instance, what does it mean to say that the graph is "composed" of subgraphs?
Please comment and discuss.
[1]: http://arxiv.org/abs/cond-mat/0503078