Network models represent complex empirical systems by means of graphs, composed of nothing more than nodes and edges, which themselves lack internal structure. Graphs can be constructed from empirical data, on the basis of simple rules that do not require much theoretical insight into the target system. Moreover, network models do not compress the data they represent. In typical network models, the mapping from data to graph is invertible. For these reasons, network modeling can seem more like a trendy format for data summary than the powerful modeling framework it is sometimes claimed to be. This chapter shows that, despite the apparent simplicity of the graph construction process, network modeling is indeed an inferentially powerful modeling framework that enables novel forms of discovery, prediction, and explanation. Thereafter, the chapter explores the fact that network properties seem to crop up repeatedly, across a wide variety of empirical domains. How surprising is this fact? Does it occur because the relevant empirical domains are intrinsically network-like, or for more pragmatic reasons to do with the way we are disposed to reason about them?