This study presents a novel approach for the analysis of the experimental dynamical behaviour of a system exhibiting Vapotron effect. This phenomenon occurs as a subcooled boiling of a refrigerant fluid entrapped in the cavities of a non-isothermally heated finned surface. A preliminary characterisation of the experimental time series has been carried out to detect the existence of a low dimensional source of the dynamics, through the adoption of non-linear time series analysis techniques. In this way the existence of chaos has been observed in the system in study, which is therefore non-linear. As a second step, a low-order non-linear model has been developed for the identification of the system dynamics. In particular, the NARMAX (Non-linear Auto-Regressive Moving Average with eXogenous inputs) identification strategy has been chosen for its flexibility, and has been implemented and generalised by means of Multilayer Perceptron neural networks. The neural model has been tested with satisfactory performances, showing the suitability of non-linear identification strategies as a reliable predictive tools for the dynamics of such kind of systems. (c) 2005 Elsevier Ltd. All rights reserved.