In this work a combination of mathematical programming based optimisation strategy and hybrid neural network models is presented in the framework of wastewater minimisation. In the optimisation strategy, the objective function is composed by three terms: freshwater cost, wastewater treatment cost and discharge taxes while the constraints are the balance equation of all the production units. Once all production units that generate wastewater and the diverse wastewater treatment systems are specified, the model automatically identifies the best treatment option for each water stream: reuse or recycle with or without regeneration. The formulated optimisation problem is solved using mathematical programming techniques and details about the optimum treatment for each stream are obtained. In general, for effluent treatment from municipal and food-industrial wastewater, the most suitable process identified is the biological treatment. Among these treatments it has been chosen the constructed wetlands. The proposed model for representing the dynamic of the wetland is based on a combination of a first principal model and an artificial neural network. The hybrid model resulted from combining both modelling strategies has been used to optimise the design of the wetland. Finally, the methodology is applied to two case studies where the characterisation of influent and effluent water flows is emphasised. © 2002 Elsevier Science B.V. All rights reserved.