In the light of current environmental concerns, pyrolysis of biomass offers a carbon neutral pathway to cheap renewable fuels known collectively as pyrolysis oil (PO). However, crude PO is not immediately usable in the current energy infrastructure and needs to be deoxygenated via upgrading technologies. Upgrading reactions are invariably complex since the chemical components in PO can run into hundreds. Moreover, these components are often very difficult to characterise, posing difficulties towards tracking their chemical reactivity and the overall kinetics as a function of time. To address this problem, the aim of this work is to present a modelling strategy to help researchers predict the kinetics of PO deoxygenation in hot compressed water, under hydrothermal conditions, near to or at the supercritical region. To do this, a trial reaction network superstructure with the maximum degrees of freedom was formulated and evaluated for the deoxygenation of three different Oils. This superstructure was based on the connectivity of an oxygen atom matrix which was quantified based on hydroxy groups by quantitative P-31{H} NMR. The complexity of the large-scale superstructure was subsequently simplified by trimming insignificant arcs; subject to an empirical understanding of the underlying chemistry. By parameter estimations, reaction networks were validated or rejected, depending on whether the computationally simulated data for a given reaction network fits the experimental results. It is anticipated that the development of the disclosed "proof of concept" models could promote the chemical understanding and hence optimization of hydrothermal upgrading technologies. (C) 2017 Elsevier Ltd. All rights reserved.