Rock avalanches are processes involving a huge amount of mobilised material, larger than 106 m3, and their propagation is characterised by a very high energy, such that no active mitigation measure is capable of protecting human lives and properties potentially exposed. Only land use planning, or emergency plans in case of occurrence of an event, can ensure safety of the communities threatened by these natural hazards. Consequently, it is necessary to predict the extent of the runout area of the process, which constitutes an objective still far from being achieved, due to the lack of knowledge about its mechanism of propagation. This PhD Thesis work mainly deals with the physical modelling of rock avalanches, which may allow for partially balancing the lack of site observations available for such processes. Laboratory tests were set-up using an unconfined inclined plane, representing the real slope, whose toe is constituted by a curved transition. A test consists in releasing a given amount of gravel, initially placed into a box (representing the source area), and measuring its propagation and final deposit features. The influence of volume, fall height, inclination of the plane and curvature radius of the transition was evaluated both on the mass propagation and on the morphology and characteristic parameters of the final deposit. In addition to the runout, width and length of the final deposit measured manually, a dedicated measurement system allowed for determining the morphology and the position of the centre of mass of the final deposit and of the mass in motion. The measurement system uses the fringe projection method, a non-destructive optical technique which allows for retrieving the shape of an object. The existing system, formerly used in a previous research work (Manzella 2008, Manzella and Labiouse 2008b), was advanced and further developed for measuring the features of the mass in motion. As the fringe projection is now performed during the whole test, the WINAnalyze code formerly used for detecting and tracking the mass front could not be used, and a new technique for measuring the front propagation was therefore developed. Numerical analyses of the laboratory tests were also carried out, using the DAN3D code (McDougall 2006, McDougall and Hungr 2004, 2005, Hungr and McDougall 2009), which assumes the mass in motion as an equivalent fluid. The code applies the depth-averaged form of St-Venant's equations, solved with a Lagrangian method base on smooth particles hydrodynamics (SPH). The basal shear resistance is modelled with simple rheological relationships. The reproduce rock avalanches, the frictional model (Coulomb’s law) and the Veollmy’s model are the most often used. Numerical modelling was performed for several tests carried out both in a previous and in this research work. The frictional rheological model is the most appropriate for reproducing laboratory tests. The range of values of the numerical parameters defined by McDougall (2006), and currently used for the back- analysis of real case studies, was modified in order to describe best the characteristics of the final deposit. The results obtained so far are promising. The DAN3D code does not present any major issue in reproducing laboratory tests characterised by different configurations of the source area and the topography.