The problems of the intermediate-range atomic structure of glasses and of the mechanism for the glass transition are approached from the low-temperature end in terms of a scenario for the atomic organization that justifies the use of an extended tunneling model. The latter is crucial for the explanation of the magnetic and compositional effects discovered in non-metallic glasses in the Kelvin and milli-Kelvin temperature range. The model relies on the existence of multi-welled local potentials for the effective tunneling particles that are a manifestation of a non-homogeneous atomic structure deriving from the established dynamical heterogeneities that characterize the supercooled liquid state. It is shown that the extended tunneling model can successfully explain a range of experiments at low temperatures, but the proposed non-homogeneous atomic structure scenario is then tested in the light of available high resolution electron microscopy imaging of the structure of some glasses and on the behaviour near the transition.