It is usually accepted in geophysics (and in civil engineering) that linear models can be used for describing an earthquake and the consequent seismic waves’ propagation. However, the large deformation experienced by the soil in these situations suggests that this paradigm requires more critical consideration. In fact, we claim that, in the vicinity of some discontinuities (that are common in all the geophysical applications of continuum models), the corresponding strain concentrations make the hypothesis of small deformation to be inadequate. In this paper, we verify the inappropriateness of the linear paradigm in a simple but reasonable case, with a view to a future application of this study to the effects of the 2009 L’Aquila earthquake. To this aim, we start with an analysis which is restricted to a two-dimensional body (i) with an inhomogeneity that resembles the Aterno River Valley, central Italy and (ii) with a non-linearity that is the most simple one, choosing the strain energy to be quadratic in the non-linear measures of deformation. More precisely, we consider a 2D piecewise homogeneous domain and a material that is viscoelastic isotropic and geometrically non-linear. We apply, to the bottom of such a domain, a seismic excitation and calculate the differences in the response between the linear and the geometrically non-linear cases. Using a suitably designed numerical model, we prove that, as conjectured, these differences not only originate near the pre-defined geometrical discontinuities but also propagate throughout the rest of the domain. Moreover, we find numerical predictions of the frequency ratios and ground acceleration time dependence and amplitude that produce, in the case of non-linear models, predictions which are closer to experimental evidence than those obtained using the corresponding linear model.