In case of exceptional floods induced by the failure of a dam, huge amounts of sediments may be eroded. This results in large-scale modifications of the valley morphology and may drastically increase the resulting damages. The objective of the research is to advance the understanding of sediment transport under dam-break flows. For such highly erosive and transient floods, it is crucial to account explicitly for sediment inertia, and therefore traditional “clear-water” modelling approaches are largely inappropriate. The present approach relies on a two-layer idealisation of the flow behaviour. Separating a clear-water flow region from the underlying sediment bed, the transported sediments are confined in a flow layer of finite thickness, endowed with its proper inertia, density and velocity. The thesis also pinpoints granular dilatancy as an essential mechanism of interaction between the layers. When passing from a solid-like to a fluid-like behaviour as they are entrained by the flow, the eroded sediment grains dilate along the vertical, and this generates vertical exchanges of mass and momentum that should be accounted for. The thesis proceeds first with experimental investigations. Laboratory dam-break waves are reproduced in a dedicated flume, exploring different bed configurations and sediment densities. Imaging observations are used to support the proposed phenomenological description of the flow. Within a shallow-water framework, theoretical and numerical endeavours are then developed to investigate the implications on the flow dynamics of the two essential contributions of the proposed description, i.e. the two-layer flow behaviour, and the effects of granular dilatancy.