We have looked for the best possible fit of the neutron-diffraction data of YBa 2 Cu 3 O 6 . 3 5 measured by Sonntag et al. using the deformation modulated structure for CuO 2 planes proposed by Krekels, Kaesche and Van Tendeloo. The resulting fit is bad. We discuss the reasons of this failure. The best explanation of the neutron data is given by the structure obtained by Aligia from the thermodynamics of oxygen ordering. Using a simple highly-correlated electronic model, we also show that there are no theoretical grounds to discard 2√2 2√2 oxygen ordered superstructures. Recent nuclear-quadrupole-resonance spectra support this conclusion.