In Chile, sand tailings dams represent the most common deposits of mining residues. These structures present a potential risk in terms of mechanical instability due to their potential susceptibility to seismic liquefaction. In order to manage these risks, it is necessary to take a probabilistic approach, thus accounting for inherent variability of material properties. However, in practice, implementing such an approach is impeded by the difficulty of acquiring and managing the data to be used in the reliability calculations and is conditioned by the relevance of the probabilistic models chosen to represent this variability. This paper proposes a method for onsite determination of the tailings relative density (DR%), and its variability, using dynamic penetration tests. This method was applied to a typical Chilean sand tailings dam, and proposes a single model for all such tailings dams by associating a probability law to the variation of DR%. Finally, the validity of this approach is demonstrated by performing a reliability calculation of liquefaction potential (which is the main cause for the failure of this type of structure in this country) for a particular tailings dam.