Abstract. This article is an introduction to the categorical theory of homotopy coherence. It is based on the construction of the homotopy coherent analogues of end and coend, extending ideas of Meyer and others. The paper aims to develop homotopy coherent analogues of many of the results of elementary category theory, in particular it handles a homotopy coherent form of the Yoneda lemma and of Kan extensions. This latter area is linked with the theory of generalised derived functors.In homotopy theory, one often needs to "do" universal algebra "up to homotopy" for instance in the theory of homotopy everything H-spaces. Universal algebra nowadays is most easily expressed in categorical terms, so this calls for a form of category theory "up to homotopy" (cf. Heller [34]).In studying the homotopy theory of compact metric spaces, there is the classically known complication that the assignment of the nerve of an open cover to a cover is only a functor up to homotopy. Thus in shape theory, [39], one does not have a very rich underlying homotopy theory. Strong shape (cf. Lisica and Mardešić, [37]) and Steenrod homotopy (cf. Edwards and Hastings [27]) have a richer theory but at the cost of much harder proofs. Shape theory has been interpreted categorically in an elegant way. This provides an overview of most aspects of the theory, as well as the basic proobject formulation that it shares withétale homotopy and the theory of derived categories. Can one perform a similar process with strong shape and hence provide tools for enriching that theory, rigidifyingétale homotopy and enriching derived categories? We note, in particular, Grothendieck's plan for the theory of derived categories, sketched out in 'Pursuing Stacks', [31], which bears an uncanny ressemblance to the view of homotopy theory put forward by Heller in [34].Grothendieck's 'Pursuing Stacks' program, in fact, again raises the spectre of doing categorical construction 'up to homotopy', as his image of a stack is as a sheaf 'up to homotopy' in which the stalks are algebraic models of homotopy types and the whole object has geometric meaning.Problems of homotopy coherence also arise naturally in studying equivariant homotopy theory. If G is a group then the equivariant homotopy of G-complexes can be studied in a useful way by translating to a category of diagrams indexed by the orbit category of G, that is the full subcategory of G-sets determined by If however G is a topological group then this category is more naturally considered to be simplicially enriched and the equivariant homotopy theory of the G-complexes ends up benefitting from results on homotopy coherence to handle the bar-resolutions etc. used to translate from the equivariant setting to the category of diagrams; see [24] and the references it contains. Again the equivariant theory looks like the discrete case but with the categorical arguments done 'up to homotopy'.In this paper we try to lay some of the foundations of such a theory of categories 'up to homotopy' or more exactly 'up to coherent ...