A backpropagation neural net was trained with the data given by a theoretical model for dynamic simulation of multistage counter-current extraction with non-miscible solvents. Despite the fact that the input fluctuation was large enough (up to 25% above and below the stationary values), the test responses of the neural model were in good agreement with theoretical model predictions. The advantage of the dynamic neural models is the extremely shore response time, and consequently they are able for process control.