Understanding the swelling behaviour of coal matrix material exposed to water vapour is of direct relevance to optimising (E)CBM recovery. However, accurate measurement of coal swelling due to water sorption presents a substantial challenge, because while measurement accuracy increases with sample size, so does equilibration time hence experiment duration. This paper reports dilatometry experiments conducted on 1 and 4 mm sized cubic samples of Brzeszcze high volatile bituminous coal. These were performed using a purpose-build, 3-dimensional (3D) dilatometer, consisting of sensitive eddy-current gap sensors. The aim was to accurately and continuously measure the volumetric response of coal matrix material during exposure to water vapour at relative humidities varied in the range of 0.1 to 95%, at a temperature of 40 degrees C. Our results show that the swelling strains attained at apparent equilibrium tend to be a factor of up to 1.45 higher perpendicular to bedding than in the bedding plane. In addition, the sample size strongly influences the swelling kinetics, but does not influence the equilibrium swelling strains. Moreover, the volumetric swelling strains attained at equilibrium show a near-linear dependence on relative humidity, reaching 137-4.43% at around 95% relative humidity. In an attempt to explain the observed behaviour, three models for swelling of unconfined coal matrix material due to water sorption were developed. These correspond to mono-layer adsorption, multiple-layer sorption, and combined mono plus multiple-layer sorption. The experimental data are equally well fitted by all three models, so that the mechanism responsible for swelling could not be uniquely identified. However, our findings do demonstrate that decreasing in situ water activity causes significant anisotropic shrinkage of coal matrix material, pointing to injection and recirculation of dry nitrogen as a promising strategy for stimulating coal seams for CBM production and as a pre-treatment for later CO2 injection and storage. (C) 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license.