Nonlinear dynamic analyses (NDAs) of embankment dams of different heights founded on a spatially variable, liquefiable alluvial layer are used to examine factors influencing embankment deformations and develop guidance on selecting representative properties for uniform analysis models. Simulations are presented for embankments ranging from 5 to
high on stochastic and uniform alluvial layers subjected to a range of input motions, with sensitivity cases including the effects of various parameters describing the alluvium and embankments. Crest settlements and slope displacements obtained from the analyses with stochastic and uniform alluvial layers are compared to obtain equivalent uniform or representative percentile properties for which a uniform model produces the same deformation as a stochastic model. The representative percentile properties to estimate median deformations from a set of stochastic realizations are generally between the 40th and 60th percentile, whereas the representative percentile properties to estimate deformations conservatively (i.e., exceeded in less than 16% of the analysis cases) are generally closer to the 30th percentile. The variability in deformation patterns obtained with the stochastic models increases as the alluvium’s scale of fluctuation in the horizontal direction increases relative to the embankment base width. Recommendations regarding factors to consider in selecting representative properties for spatially variable alluvial foundations in NDAs of embankment dams and the corresponding variability in deformations are presented.