The risk of overfitting pair distribution function (PDF) data for highly defective material (Farrow et al., 2007) is illuminated with the example of the nanocrystalline hydrous ferric oxyhydroxide, ferrihydrite. Two structural models have been published by Michel et al. (2007, 2010) using this method, both of which contradict the standard `ferrihydrits' model established by X-ray diffraction (Drits et al., 1993), and confirmed by single-crystal electron nanodiffraction (Janney et al., 2001) and neutron diffraction (Jansen et al., 2002). Although PDF data are reproduced equally well with the two regression models, neither model is realistic: the first (fhyd6) violates Pauling's 2nd rule, and the second (ferrifh), Pauling's 3rd rule.