The problems of mathematical modeling of complex systems are considered (in the present paper) in terms of neural network technique. The construction of robust neural network model of processes in porous catalyst is presented as an example. The results of neurocomputing are performed.