Abstract. The vanishing of the divergence of the matter stress-energy tensor for General Relativity is a particular case of a general identity, which follows from the covariance of the matter Lagrangian in much the same way as (generalized) Bianchi identities follow from the covariance of the purely gravitational Lagrangian. This identity, holding for any covariant theory of gravitating matter, relates the divergence of the stress tensor with a combination of the field equations and their derivatives. One could thus wonder if, according to a recent suggestion [1], the energy-momentum tensor for gravitating fields can be computed through a suitable rearrangement of the matter field equations, without relying on the variational definition. We show that this can be done only in particular cases, while in general it leads to ambiguities and possibly to wrong results. Moreover, in nontrivial cases the computations turn out to be more difficult than the standard variational technique.In a recent paper [1] Accioly et al. have observed that for some well known cases of classical fields interacting with gravitation (e.g. scalar field, electromagnetic field), upon contracting the dynamical equation for the matter field with a suitable linear combination of covariant derivatives of this field, the resulting expression represents the vanishing of the covariant 4-divergence of some rank-two tensor. In the cases considered in [1], the latter turns out to coincide exactly with the stress-energy tensor of the matter field, according to the usual variational definition.Let us explain the geometrical origin of this phenomenon. For the reader's convenience, we recall the derivation of the strong conservation law (sometimes called "Bianchi identity for matter") in the case of metric theories of gravity, in the form suitable for our aim. We denote arbitrary matter fields by ψ A ; the signature of the metric g µν is chosen to be (− + ++); the expressions ∇ µ f and f ;µ both denote covariant derivation relative to the metric g µν and we set c = 8πG = 1. In accordance with [1], we assume that pure gravitation is described by the usual Einstein-Hilbert Lagrangian, although the considerations below hold, mutatis mutandis, under more general assumptions. The matter Lagrangian is L(g, ψ) = L(g µν , R λ µνσ , ψ A , ψ A;µ ): the dependence on g includes a possible non-minimal coupling to the curvature. The resulting action is S = Ω