Abstract : Known in its principle in the early 1930s, the belt grinding process especially came out industrially at the American car manufacturer's Chrysler. He was the first to develop an abrasive belt set-up allowing a mechanised operation for a process of manufacturing merely used for off-hand grinding. The superfinishing by coated belts won then a broad audience under the impulsion of the works of German Karl Weick in 1950. The objective of this PhD thesis was to specify the action of belt grinding in respect to the involvement of the working variables as well as the main characteristics of physical mechanisms which are activated in the superfinishing of metals. In summarised open literature, we showed the insufficiency of scientific works on the grinding by coated belts which bring back the knowledge of this abrasive technology to a bushel of empirical facts showing the critical involvement of a certain number of working variables of the process. To better understand the physical mechanisms which govern the machining by abrasive film, we introduced a tribo-energetic approach which match respectively the in situ physical measurements during the belt grinding experiments and the post-test analyses of the workpiece and the coated belt. Taking then methodically the knowledge of design of belt grinding technologies, we developed an original test bench based on a grinding test with abrasive belt where all limit parameters are given by the industrial conditions (plant test). The activated physical mechanisms in the belt grinding of the machined material and at the level of the coated belt / workpiece interface were then discussed. Also the roles of the working process variables examined with a scale effect were specified. Although the grinding with abrasive belts operates by micro-shaving of roughness and by loading of micro-chips; the structure of the coated film, characterized by a high power of accommodation (micro-chip build-up), seems to slow down or even transform this action of cutting into a phenomenon of rubbing and ploughing. Moreover, ductility and micro-structural homogeneity of the machined material seems to control, between other phenomena, the thickness of workhardened subsurface layer and the endurance of the belt finished surface. We also presented original results which show, by a multiscale approach of surface topography, the scale effect of roughness on the roles of working variables of the process. This one almost determines the process stability when grinding by abrasive belts and better allows specifying the tribological quality of surface in terms of functional needs.