N-doped brookite-rutile catalysts were prepared using the sol-gel method with different nitrogen precursors, namely urea (CH4N2O), propionitrile (C3H5N), ammonium hydroxide (NH4OH), ammonium nitrate (NH4NO3) and ethylene diamine (C2H8N2). Testing the photoactivity of the prepared catalysts with 4-nitrophenol revealed that ammonium nitrate was the best doping agent with a nominal N-content of 0.8% (w/w), yielding a 5-fold increase in the pseudo-first order constant of 4-nitrophenol disappearance and a 3-fold increase in the pseudo-first order constant of TOC disappearance, with respect to undoped TiO2, when irradiated with LED visible light (>425 nm). In the same experimental conditions, commercial catalysts such as Evonik P25 and mixtures of commercial rutile and brookite failed to work. XRD allowed to identify two crystal phases, i.e. brookite and rutile, and to show that the most active catalyst had the highest brookite and lowest amorphous content, along with the largest rutile crystallites. Through HRTEM, the morphology and crystallinity were further investigated: brookite particles were much smaller and roundish with respect to rutile and the intimate contact between the two phases was also well highlighted. N-doping did not produce oxygen vacancies as shown by Raman spectroscopy; thus, the doping can be considered interstitial rather than substitutional. Surface hydroxylation did not promote oxidation ability, as revealed by TGA-DTA: the most reacting catalyst is the least hydroxylated one. BET revealed that the samples are partially mesoporous (type IV hysteresis), although no template/surfactant was used, and the pore size and volume seemed to affect their activity. UV-vis DRS allowed to extrapolate the band gaps, only slightly narrower for N-doped samples, which, however, showed a pronounced absorption of visible radiation compared to undoped TiO2. Photoluminescence showed that the emission due to electron-hole recombination decreases with the N-loading, eventually reaching a minimum plateau for doping amounts just above the optimal one. (C) 2016 Elsevier Inc. All rights reserved.