Detailed investigations have been carried out to check the ability of multilayer neural networks to model the simultaneous mass transfer and chemical reaction in the liquid-liquid reacting system. In this approach the intrinsic reaction kinetics and diffusive mass transfer are represented by a black-box and only the input-output signals are analysed. The data for training of the net have been taken from the experiments performed in a RC1 Mettler Toledo reaction calorimeter. The hydrolysis of propionic anhydrite catalysed with sulphuric acid has been chosen as a testing reaction. The hybrid, first-principle-neural-network model has been defined to describe batch and semibatch stirred tank reactors operating at different conditions. Good accuracy and flexibility of the proposed approach have been obtained for a properly defined experimental programme supplying data for learning.