The graph state formalism is a useful abstraction of entanglement. It is used in some multipartite purification schemes and it adequately represents universal resources for measurement-only quantum computation. We focus in this paper on the complexity of graph state preparation. We consider the number of ancillary qubits, the size of the primitive operators, and the duration of preparation. For each lexicographic order over these parameters we give upper and lower bounds for the complexity of graph state preparation. The first part motivates our work and introduces basic notions and notations for the study of graph states. Then we study some graph properties of graph states, characterizing their minimal degree by local unitary transformations, we propose an algorithm to reduce the degree of a graph state, and show the relationship with Sutner sigma-game. These properties are used in the last part, where algorithms and lower bounds for each lexicographic order over the considered parameters are presented.