BACKGROUND: TiO2 heterogeneous photocatalysis should be optimized before application for the removal of pollutants in treated wastewaters. The response surface methodology (RSM) and artificial neural networks (ANNs) were applied to model and optimize the photocatalytic degradation of total phenolic (TPh) compounds in real secondary and tertiary treated municipal wastewaters. RESULTS: RSM was developed by considering a central composite design (CCD) with three input variables, i.e. TiO2 mass, initial concentration of TPh and irradiation intensity. At the same time a feed-forward multilayered perceptron ANN trained using back propagation algorithms was used and compared with RSM. Under the optimum conditions established in experiments ([TPh]0 = 3 mg L-1; [TiO2] = 300 mg L-1; I = 600 W m-2) the degradation for both TPh and total organic carbon (TOC) followed pseudo-first-order kinetic model. Complete degradation of TPh took place in 180 min and reduction of TOC reached 80%. A significant abatement of the overall toxicity was accomplished as revealed by Microtox bioassay. CONCLUSIONS: It was found that the variables considered have important effects on TPh removal efficiency. The results demonstrated that the use of experimental design strategy is indispensable for successful investigation and adequate modeling of the process and that ANNs gave better modelling capability than RSM. Copyright (c) 2012 Society of Chemical Industry