In view of some recent works on the role of vertex corrections in the electron-phonon system we readress an important question of the validity of the Migdal-Eliashberg theory. Based on the solution of the Holstein model and 1/λ strongcoupling expansion, we argue that the standard FeynmanDyson perturbation theory by Migdal and Eliashberg with or without vertex corrections cannot be applied if the electronphonon coupling is strong (λ ≥ 1) at any ratio of the phonon, ω and Fermi, EF energies. In the extreme adiabatic limit (ω << EF ) of the Holstein model electrons collapse into self-trapped small polarons or bipolarons due to spontaneous translational-symmetry breaking at λ ≃ 1. With the increasing phonon frequency the region of the applicability of the theory shrinks to lower values of λ < 1.PACS numbers: 74.65.+n,74.60.Mj The theory of ordinary metals is based on 'Migdal's' theorem [1], which showed that the contribution of the diagrams with 'crossing' phonon lines (so called 'vertex' corrections) is small. This is true if the parameter λω/E F is small. It was then shown that if the adiabatic BornOppenheimer approach is properly applied to a metal, there is only negligible renormalisation of the phonon frequencies of the order of the adiabatic ratio, ω/E F << 1 for any value of λ [3]. The conclusion was that the standard electron-phonon interaction could be applied to electrons for any value of λ but it should not be applied to renormalise phonons [4]. As a result, many authors used the Migdal-Eliashberg theory with λ much larger than 1 (see, for example, Ref.[5]).However, starting from the infinite coupling limit, λ = ∞ and applying the inverse (1/λ) expansion technique [6] we showed [7][8][9] that the many-electron system collapses into small polaron regime at λ ∼ 1 almost independent of the adiabatic ratio. In this letter I compare the Migdal solution of the Holstein Hamiltonian with the exact one in the extreme adiabatic regime, ω/E F → 0, to show that the ground state of the system is a self-trapped insulating state with broken translational symmetry already at λ ≥ 1.The vertex corrections and the finite bandwidth are rather technical issues, playing no role in the extreme adiabatic limit [1,32,31]. There is, however, another basic assumption of the canonical Migdal-Eliashberg theory. That is the electron and phonon Green functions (GF) are translationally invariant. As a result one takes G(r, r ′ , τ ) = G(r− r ′ , τ ) with Fourier component G(k, Ω) prior to solving the Dyson equations. This assumption excludes the possibility of local violation of the translational symmetry due to the lattice deformation in any order of the Feynman-Dyson perturbation theory. This is similar to the absence of the anomalous (Bogoliubov) averages in any order of the perturbation theory. To enable electrons to relax into the lowest polaronic states, one has to introduce an infinitesimal translationally noninvariant potential, which should be set equal to zero only in the final solution for the GF [8]. As in the case of the...