Abstract Among the four dominant mechanisms of internal erosion, suffusion appears as one of the most complex. It is indeed the result of the combination of three processes: dislodgement of fine particles, transport of them, and filtration of some fluidized fine particles. These processes depend on the soil’s stress state and on the hydraulic gradient path. Thus, to ensure the repeatability of the suffusion test and to study with accuracy the influence of the aforementioned parameters, a new apparatus was developed. This apparatus is designed to test specimens under a vertical downward flow in hydraulic gradient controlled condition while controlling the confining pressure and the stress deviator, which can be either positive or negative. Ten tests on one cohesionless soil were performed in triaxial conditions, with the same mean effective stress and four different values of stress deviator. To compare with conventional suffusion results, one test is also performed in rigid wall conditions. For all performed tests, the same hydraulic gradient path was applied and particular attention is paid to the repeatability, which implies control of each test step. At the end of each test, the specimen was divided into four layers to measure post-suffusion grain size distributions. The time evolutions of the hydraulic conductivity and the erosion rate permit to identify four different phases. Each phase is characterized by two methods: one based on the hydraulic gradient and the second based on the expended energy by the fluid. The results show the great influence on the suffusion kinetics of the preferential flow paths, which localize in the body of the specimen in triaxial conditions and on the circumference in rigid wall ones. Under a constant mean effective stress, the effect of the stress deviator on the suffusion kinetics appears limited, for the tested soil and shear stress ratios.