We obtained improved optical positions for 300 ICRF2 sources - the Rio survey. We compared the Rio survey with 10 other selected optical astrometric surveys and studied the link between the Hipparcos Catalogue Reference Frame (HCRF) and the International Celestial Reference Frame, Second Realization (ICRF2). We investigated the possible causes for the observed non-coincidence between the optical and ICRF2 positions. The Rio survey positions were referred to the second version of the United States Naval Observatory CCD Astrograph Catalog (UCAC2), currently the best tested HCRF densification. The sources are between -90{deg}<{delta}<+30{deg}. We used two telescopes with suitable diameters and focal lengths to properly link the observed ICRF2 sources with the UCAC2, using intermediate brightness stars. We certified the astrometry done with many statistical tests. The average 'optical minus ICRF2' offsets and respective standard deviations in ({alpha}, {delta}) were -3mas (41mas) and +4 (45mas). The Rio survey represents well the zero-point offset of the other surveys. The standard error of 3.5mas found for the HCRF/ICRF2 link indicates an error excess that can be originated by a non-coincidence between the observed optical/VLBI positions. We thus discussed the influence of the errors from the UCAC2. Then, we searched for correlations with the source morphology, represented by structure indices defined in the radio and in the optical domain. Finally, we studied how the position offsets could originate from the perturbation of the optical point spread function (PSF) of the source's core, by a second source of flux. We found an analytical relation that describes the resulting centroid shift, as a function of the atmospheric seeing, the brightness ratio and the relative distance between the two contributing flux sources. Two scenarios, modelled by this relation, are discussed: an extinction window in the dust torus nearby the core, and a Galactic star near the line of sight.