This paper presents a neural network based approach to the identification and control of an experimental natural circulation loop. The aim of the model is to predict the dynamical evolution of the oscillations, characterizing the system dynamics in some operating conditions and that can cause dangerous flow reversal. The identification of the system was the first step towards the design of an appropriate control system, which was then addressed again using a neural network. The neural approach herein proposed is based on a cascade of several neural networks representing both the system and the controller. As a first step the neural modeling of the system performed by using a suitable set of experimental data. In the second step, a neural control is trained, through a suitable learning algorithm, in order to obtain the desired behaviour of the system.