Today's state-of-the-art method for calculating the interaction of atoms or small molecules with metal surfaces is considered to be density functional theory (DFT) at the generalized gradient approximation (GGA) level employing a slab or supercell representation of the surface. The method is widely used and by many assumed to be both qualitatively and quantitatively accurate. This notion has recently been challenged by Feibelman [J. Phys. Chem. B 105, 4018 (2001)] who suggest that the DFT/GGA method does not correctly predict the most stable adsorption site for the CO/Pt(111) system, and they conclude that the method is not qualitatively accurate. However, using a different calculational approach we find a good agreement between the calculated potential energy surface for this system and the one inferred from experiments, indicating that the evidence supporting the view of Feibelman is not yet conclusive. On the contrary, we advocate the view that the DFT/GGA method should at the moment be considered qualitatively accurate for predicting the most stable CO adsorption sites on metal surfaces. This view is supported by (i) our results for the Pt(111) surface which in agreement with experiments favors the top site, (ii) an assessment of literature results for other surfaces, suggesting that the error in the relative stability of the CO adsorption sites on a given surface is within +/-0.1 eV when compared to experiments, (iii) the considerable challenge one faces when trying to converge DFT/GGA calculations within +/-0.1 eV with respect to all computational parameters, (iv) and that for energy differences smaller than say 0.1 eV, calculated quantities like, e.g., vibrational frequencies and geometries discriminate correctly between sites, being in agreement with experiments at the correct adsorption site. (C) 2003 American Institute of Physics.