Backward erosion piping is an internal erosion mechanism during which shallow pipes are formed in the direction opposite to the flow underneath water-retaining structures as a result of the gradual removal of sandy material by the action of water. It is an important failure mechanism in both dikes and dams where sandy layers are covered by a cohesive layer. Sand boils can indicate that backward erosion is present and they are observed regularly during high water and floods. Although failure resulting from backward erosion piping is not common, several dike failures in the US, China and the Netherlands have been attributed to this mechanism. Given the impact that climate change is expected to have, prediction models for backward erosion piping are becoming increasingly important in flood-risk assessment. The prediction models available until now, such as Bligh’s rule and the Sellmeijer model, were validated in the research programme ‘Strength and loads on flood defence structures’ (SBW: Sterkte en Belastingen Waterkeringen) in the period 2008-2010 using small-, medium- and large-scale experiments. These experiments showed that an empirical adjustment of the Sellmeijer model was required to take the effect of the sand type into account and that validation was not possible for loose sand types because the erosion mode is different in those conditions. However, the absence of a theoretical basis makes this proposed empirical adjustment unsatisfactory because it lacks robustness. The main question addressed by this dissertation is how to explain and predict the pipe-forming erosion processes in uniform sands. A review of the literature, in conjunction with additional experiments, showed that the critical head in pipe formation leading to dike failure depends on either pipe initiation or pipe progression. In some experiments, the critical head for pipe initiation exceeds that of pipe progression and equilibrium is therefore prevented. The experiments in which no equilibrium was observed allowed for the development of a model for pipe initiation. It was possible to relate the observed differences in critical gradient caused by scale, sand type and configuration to the fluidisation of sand very close to the exit, where the local gradients are high. In the field, pipe progression is likely to determine the critical gradient. The Sellmeijer model predicts the progression of the pipe on the basis of the equilibrium of particles on the bottom of the pipe. The literature, and an analysis of the pipe width, depth, gradient and erosion process in experiments, indicate that pipe progression relies on two processes: primary erosion, which causes the removal of particles at the pipe tip, and secondary erosion, which causes the erosion of the pipe walls and bottom. Although the Sellmeijer model does not include primary erosion, it does function well for sand layers with a 2D exit configuration in which there is no variation in the grain size along the pipe path. The adaptation of the Sellmeijer model that was found necessary to account for the effect of sand type can be replaced by using the original model in combination with a variable bedding angle based on incipient motion experiments from the literature. The Sellmeijer model does not predict the critical gradient well for 3D configurations such as flow towards a single point, or for heterogeneous soils. Variations in the grain size in the pipe path were found to result in significantly higher critical gradients than expected, whereas a strong concentration of the flow towards the exit led to a fall in the critical gradient. 3D numerical calculations and the inclusion of primary erosion in the Sellmeijer model are needed to predict piping under these conditions.