We develop a method for spatio-spectral localization of harmonic data on a sphere and use it to interpret recent high-resolution global estimates of the gravity and topography of Venus in the context of geodynamical models. Our approach applies equally to the simple spatial windowing of harmonic data and to variable-length-scale analyses, which are analogous to a wavelet transform in the Cartesian domain. Using the variable-length-scale approach, we calculate the localized RMS amplitudes of gravity and topography, as well as the spectral admittance between the two fields, as functions of position and wavelength. The observed admittances over 10 per cent of the surface of Venus (highland plateaus and tessera regions) are consistent with isostatic compensation of topography by variations in crustal thickness, while admittances over the remaining 90 per cent of the surface (rises, plains and lowlands) indicate that long-wavelength topography is dominantly the result of vertical convective tractions at the base of the lithosphere. The global average crustal thickness is less than 30 km, but can reach values as large as 40 km beneath tesserae and highland plateaus. We also note that an Earth-like radial viscosity structure cannot be rejected by the gravity and topography data and that, without a mechanical model of the lithosphere, admittance values cannot constrain the thickness of the thermal boundary layer of Venus. Modelling the lithosphere as a thin elastic plate indicates that at the time of formation of relief in highland plateaus and tesserae, the effective elastic plate thickness, Te, was less than 20 km. Estimates of Te at highland rises are consistently less than 30 km. Our inability to find regions with Te > 30 km is inconsistent with predictions made by a class of catastrophic resurfacing models.