Full-waveform inversion (FWI) is not yet a mature imaging technology for lithospheric imaging from teleseismic data. Therefore, its promise and pitfalls need to be assessed more accurately according to the specifications of teleseismic experiments. Three important issues are related to (1) the choice of the lithospheric parametrization for optimization and visualization, (2) the initial model and (3) the acquisition design, in particular in terms of receiver spread and sampling. These three issues are investigated with a realistic synthetic example inspired by the CIFALPS experiment in the Western Alps. Isotropic elastic FWI is implemented with an adjoint-state formalism and aims to update three parameter classes by minimization of a classical least-squares difference-based misfit function. Three different subsurface parametrizations, combining density (ρ) with P and S wave speeds (Vp and Vs) , P and S impedances (Ip and Is), or elastic moduli (λ and μ) are first discussed based on their radiation patterns before their assessment by FWI. We conclude that the (ρ, λ, μ) parametrization provides the FWI models that best correlate with the true ones after recombining a posteriori the (ρ, λ, μ) optimization parameters into Ip and Is. Owing to the low frequency content of teleseismic data, 1-D reference global models as PREM provide sufficiently accurate initial models for FWI after smoothing that is necessary to remove the imprint of the layering. Two kinds of station deployments are assessed: coarse areal geometry versus dense linear one. We unambiguously conclude that a coarse areal geometry should be favoured as it dramatically increases the penetration in depth of the imaging as well as the horizontal resolution. This results because the areal geometry significantly increases local wavenumber coverage, through a broader sampling of the scattering and dip angles, compared to a linear deployment.