Previous works in batch reactors have shown for activated carbon cloths quick mass transfer and high adsorption capacities against organic molecules present in water. In this study, activated carbon cloth is set up in a dynamic reactor to determine its adsorption performance for various micropollutants and operating conditions. Experimental data of breakthrough curves for different conditions are generated using a pilot-unit to determine breakthrough times and adsorption capacities. Classical models set up for porous materials are not really useful for adsorption on activated carbon cloth filter. A neural network approach is then carried out in order to model experimental data by taking into account specific characteristics of cloths as input neurons. Equilibrium and kinetic parameters are also comprised in the input layer (pore diffusion coefficients and Freundlich constants). The relative influence of each input variable on the outlet concentration is analysed. Calculated and experimental data are compared, good predictions are obtained using this approach.