In this paper, the reliability of the characterization of thin oxides at the nanoscale using Scanning Capacitance Spectroscopy (SCS) is addressed. 5 nm thick thermal silicon dioxide layers are probed by SCS and compared to capacitance versus voltage (C-V)curves obtained with a classical ("macroscopic") bench using large (e.g. 100 mu m x 100 mu m) electrodes. The detrimental influence of the laser used for the measurement of the deflection of the AFM cantilever is pointed out. It is shown that SCS is not able to provide the correct shape of the dC/dV versus voltage curve, compared to macroscopic C-V curves. In particular, a hysteretic behaviour and an increase of the signal in the inversion region are measured by SCS whereas they are not present in macroscopic C-V curves. We show that this can be explained by a bad electrical contact (presence of contaminants and/or charges) between the Atomic Force Microscope's tip and the oxide. Macroscopic C-V curves obtained directly with a large metallic tip positioned on the oxide layer show the same features as SCS, while SCS operated on an electrode cut from an electrode of macroscopic size used for macroscopic C-V curves leads to the correct shape. This shows that SCS won't be fully comparable to macroscopic measurements unless the influence of the contact between the AFM tip and the sample is taken into account. (C) 2009 Elsevier B.V. All rights reserved.