We perform a detailed analysis of all existing data (CELLO, CLEO, BABAR) on the pion-photon transition form factor by means of light-cone sum rules in which we include the next-to-leading order QCD radiative corrections and the twist-four contributions. The next-to-next-to-leading order radiative correction together with the twist-six contribution are also taken into account in terms of theoretical uncertainties. Keeping only the first two Gegenbauer coefficients a 2 and a 4 , we show that the 1 error ellipse of all data up to 9 GeV 2 greatly overlaps with the set of pion distribution amplitudes obtained from nonlocal QCD sum rules-within the range of uncertainties due to twist-four. This remains valid also for the projection of the 1 error ellipsoid on the ða 2 ; a 4 Þ plane when including a 6 . We argue that it is not possible to accommodate the high-Q 2 tail of the BABAR data with the same accuracy, despite opposite claims by other authors, and conclude that the BABAR data still pose a challenge to QCD.