In this paper we develop homotopy theoretical methods for studying diagrams. In particular we explain how to construct homotopy colimits and limits in an arbitrary model category. The key concept we introduce is that of a model approximation. A model approximation of a category $mathcal{C}$ with a given class of weak equivalences is a model category $mathcal{M}$ together with a pair of adjoint functors $mathcal{M} rightleftarrows mathcal{C}$ which satisfy certain properties. Our key result says that if $mathcal{C}$ admits a model approximation then so does the functor category $Fun(I, mathcal{C})$. From the homotopy theoretical point of view categories with model approximations have similar properties to those of model categories.They admit homotopy categories (localizations with respect to weak equivalences). They also can be used to construct derived functors by taking the analogs of fibrant and cofibrant replacements. A category with weak equivalences can have several useful model approximations. We take advantage of this possibility and in each situation choose one that suits our needs. In this way we prove all the fundamental properties of the homotopy colimit and limit: Fubini Theorem (the homotopy colimit - respectively limit- commutes with itself), Thomason's theorem about diagrams indexed by Grothendieck constructions, and cofinality statements. Since the model approximations we present here consist of certain functors 'indexed by spaces', the key role in all our arguments is played by the geometric nature of the indexing categories.