The stability analysis of earth dams faces numerous sources of uncertainty due to the spatial variability of the physical (wet and saturated density) and mechanical (cohesion, internal friction angle) properties of the embankment constituting the different dike layers. This variability is linked firstly to the natural heterogeneity of their properties depending on the position in the plan and in-depth at their borrowing zone, as well as to the process of their implementation of inhomogeneous layers. In this research, the spatial variability of these uncertain parameters is randomly generated by the Karhunen-Loève discretization method. This method is applied to a practical case of a homogeneous dike made of local material. The auto-covariance model, as well as the distribution laws adapted to these parameters, are determined from two samples actually measured on the backfill of the body of an earthen dike, these are the dry density as well as the content in water. Statistical analysis is carried out in order to adjust distribution laws validated by adequacy tests, such as Kovolmogor_Smirnov. Random generation maps of the different physical and mechanical parameters are proposed at any point of the dike body, thus reflecting the uncertainty recorded by the measurement samples.