Membrane bioreactor systems are increasingly applied for municipal wastewater treatment, largely as submerged membrane units with rather low transmembrane pressure difference and feed-sided air pulsing. The combination of activated sludge units and membrane filtration for biomass retention generally results in high effluent quality and compact plant configuration. This paper proposes two permeate flux models for the filtration process in membrane bioreactor (MBR) systems. One semi-empirical model describes filtration resistance based on operational parameters and a hypothetical membrane age. The membrane age reflects performance-determining processes evolving on long time scales. The model comprises a set of model parameters which were calibrated using data from a pilot scale MBR operating with PURON submersed hollow fibre membrane modules. The second model is more empirical in nature and builds on an artificial neural network (ANN). The training procedure for the ANN was conducted based upon pilot-studies with an MBR system using a novel submerged capillary module supplied by PURON. Good correlations were found between the predicted and measured permeability using both models.