We introduce a framework called graph-rewriting automata to model evolution processes of networks. It is a natural extension of cellular automata in the sense that a fixed lattice space of cellular automata is extended to a dynamic graph structure by introducing local graph-rewriting rules. We consider three different constructions of rule sets to show that various network evolution is possible: hand-coding, evolutionary generation, and exhaustive search. Graph-rewriting automata provide a new tool to describe various complex systems and to approach many scientific problems.