The paper presents a collaborative piece of research undertaken by seven research teams from different universities within the MUSE network. The objective is to benchmark different approaches to constitutive modelling of mechanical and water retention behaviour of unsaturated soils by comparing simulations of suction controlled and constant water content laboratory tests. A set of thirteen triaxial and oedometer laboratory tests, covering the mechanical and water retention behaviour of an unsaturated compacted silty soil under a variety of stress paths, has been provided by one of the seven participating teams. This data set has then been used by the other six teams for calibrating a constitutive model of their choice, which has been subsequently employed for predicting strains and degree of saturation in three of the thirteen tests used for calibration as well as in one "blind" test for which experimental results had not been previously disclosed. By comparing predictions from all teams among themselves and against experimental data, the work highlights the capability of some of the current modelling approaches to reproduce specific features of the mechanical and water retention behaviour of unsaturated soils helping to identify potential areas of weakness where future research should focus. It also demonstrates the dispersion of results to be expected when different constitutive models, independently calibrated by different teams of researchers, are used to predict soil behaviour along the same stress path.