We search for deviations from next-to-leading order QCD evolution in HERA structure function data. We compare to data predictions for structure functions in the small x region, obtained by evolving backwards to low Q^2 the results of a parton fit performed in the large Q^2 region, where fixed-order perturbative QCD is certainly reliable. We find evidence for deviations which are qualitatively consistent with the behaviour predicted by small x perturbative resummation, and possibly also by nonlinear evolution effects, but incompatible with next-to-next-to leading order corrections.
Bibliographical noteFinal version (to be published in PLB), new plots in fig. 5 and new column in tab. 4 added, fig. 10 added, discussion on distances on pag. 8 added, typos corrected