Although the mining residues in a global way can be classified in a single class of granular material, a set of causes lead to a variability of the geotechnical characteristics of the material and of the internal structure of the dams constituted from these. Consequently, to adopt a hypothesis of homogeneity to estimate the mechanical stability of this kind of works (slope stability, liquefaction risk,..), can drive to a not realistic evaluation of the safety. To study the risk of stability of tailings dams, it is necessary to have a probabilistic approach. Indeed, the variability of the soils properties is the main cause of the uncertainty in the geotechnical design. It is thus necessary to know the distribution laws and the characteristics of average, deviation or distribution of the input parameters of soil mechanical models. The obtaining of these laws by laboratory test is boring, expensive and long. This article suggests demonstrating that it is possible to estimate these input parameters and them characteristics of distribution from the measure of the variability on the physical and state characteristics by means of in situ tests. By this way, authors show it is possible to obtain a very good estimation of the friction angle and carry out a probabilistic study of dam stability taking into account the mechanical parameters variability.