The Random Finite Element Method (RFEM) has been shown in many theoretical publications to offer advantages in the quantification of the probability of failure. However, it has rarely been applied in real situations (geometry, material properties, soil layers) and seldom, if at all, to a well instrumented geotechnical failure. This paper reports a case study of a full-scale controlled dyke failure, where the heterogeneity was previously measured via CPTs (Cone Penetration Tests), and the dyke itself was highly instrumented. This offers the opportunity to compare and apply various techniques previously developed (e.g. random field conditioning) with field data, rather than to computer generated data. The RFEM analyses presented are compared with deterministic analyses, demonstrating the relative performance of the methods.@en