This is just to give different point of view to the problem. 1. Suppose we have a category C and a full subcategory H (think of H as a subcategory of injective objects). Suppose we have a functor E:C --> H together with natural transformation i: Id --> E(C) such that i is monic and ,moreover, E is weak left adjoint to the inclusion functor H --> C. Then I claim that under two additional conditions E is genuine adjoint and i is unit of the adjunction. The conditions are: a. the natural transformation i is identity on H. b. E preseves monics. The proof is just a repetition of P.Freyd proof. We have to prove that the extension in the diagram i:A --> E(A) | / f | / | / I is unique for any morphism f and "injective" I. By applying E to this diagram and using condition a) we see that it is sufficient to prove that E(i) is identity. Repeating Freyd's proof we see that E(i) is idempotent. Using condition b) we conclude that it is identity. The conditions a) and b) are obviously satisfied in the case when E is "injective hull functor" (of coarse a) is true up to iso, again see P,Freyd proof). As the unit of the adjunction is monic so the functor E reflects epimorphisms. If in C we have that mono + epi implies iso we finally have that i is iso and ,hence, the result of Adamek, Herrlish, Rosicky and Tholen. 2. I would not dare to simply repeat P.Freyd argument if I don't have another proof (in a special but important case). The proof is much more techniqual but I believe reflects another interesting side of the problem of functoriality of injective resolutions. Consider the following bicategory. The objects are categories enriched in the closed monoidal category of (say bounded ) chain complexes. The 1-arrows are enriched distributors. The 2-arrows are homotopy classes of "coherent" natural transformations (i.e. we localize the category of natural transformations with respect to the class of morphisms which are level quaziisomorphisms). We have to define the composition of 1-arrows. This requires some techniques but in a few words the result is left derived functor of the composition of enriched distributors. The resulting bicategory is closed on the left and right. Now, for a chain functor K: A --> B we can consider the right Kan extension of it along itself in the above bicategory (codensity monad). The Kleisli category of it is called "strong shape theory of K" Ssh_K. It is possible to prove, that: c). If K is a full embedding and has an enriched left adjoint then Ssh_K is just a Kleisli category of the corresponding monad on B. d). If B is the category of chain complexes (say bounded)in an abelian category with enough injectives and A is full subcategory of injective chain complexes, then homology of Ssh_K(X,Y) are isomorphic to the right derived functor of internal Hom of B. In particular, if X and Y are concentrated in the dimension 0 the cohomology are just the correspobding Ext's. Coming back to the original problem. If H\in C are abelian and satisfy the conditions a),b) then, according to our calculations the inclusion K: Ch(H) ---> Ch(C) has a right adjoint and , hence,(by point c) Ssh_K(X,Y) = Hom(X,E(Y)) for X,Y concentrated in dimension 0. Hence by d) the injective dimension of C is 0 and we have the result again. Another words the injective dimension is the obstruction for nonnaturality of the injective hull. The same sort of theory can be developed for simplicial (or Cat) enriched situation see Batanin.M, Categorical Strong Shape THeory, Cahiers de Topologie et Geom, v,XXXVIII-1(1997)p. 3-66. I think some other results of nonaturality (as , for example, the result of Shakhmatov mentioned by AHRT on p.8) are related to these strong shape categories. Michael Batanin.