Networks are a common way of representing linked information. The goal of network abstraction is to transform a large network into a smaller one, so that the smaller is a useful summary of the original graph. In this paper we review dierent approaches and techniques proposed to abstract a large network. We classify the approaches along two axes. The rst one consists of elementary simplication techniques used: pruning of (irrelevant) nodes and edges, partitioning to several smaller networks, and generalization by replacement of subnetworks by more general structures. The other axis is objective vs. subjective methods; the latter ones aim to maintain more information about those parts of a network that the user has indicated as interesting. We conclude the review by a brief analysis of which intersections of the two axes are least researched and could therefore have future potential.