Artificial neural networks (ANNs) were used to estimate from online pH measurements the ammonium concentration in an anaerobic ammonium oxidation (ANAMMOX) sequencing batch reactor (SBR) treating reject water (RW) from the anaerobic treatment of domestic wastewater. The SBR was initially fed with a synthetic autotrophic medium (SM) to assure a stable and ANAMMOX dominating process. After the SBR had been operating stable for 1 month, the removal efficiencies of ammonium and nitrite were equal to 9122 +/- 3.92% and 94.16 +/- 8.76%, respectively. The experimental data obtained in this period was taken as basis but not used directly for the training of the ANNs. Instead, the data was used for the calibration of an ordinary differential equations (ODE) model implemented to simulate the nitrogen removal processes that took place in the SBR. This action helped to increase the amount of available data, thereby improving the teaming capacity of the networks and reducing the need of extensive experimental analysis. After parameter calibration, the experimental data agreed well with the simulation results in the case of ammonium and nitrite. The simulated ammonium concentration (broadened data set) was then used as target data for the training of different structures of two types of ANNs: multilayer feedforward neural network (MLFNN) and adaptive-network-based fuzzy inference system (ANFIS). The ANNs structures with the best performance after training yielded correlation coefficients (R) of R-MLFNN=0.9924 and R-ANFIS=0.9922. Afterwards, the selected ANNs were validated by comparing the predicted ammonium concentration with the experimental values obtained during the adaptation from SM to the targeted RW. Both types of ANNs were able to predict with good accuracy the ammonium removal inside the SBR even while dealing with the largely fluctuating influent conditions without the need of further training. The results obtained after validation were R-MFLNN=0.8440 and R-ANFIS=0.8454. This shows the potential that ANNs have to model the ANAMMOX process if enough and representative data is available for training. (C) 2015 Elsevier Ltd. All rights reserved.