Soil erosion due to rainfall is a complex phenomenon. Factors influencing erosion and subsequent transport include rainfall intensity, soil properties, topography, land cover and antecedent conditions. These factors and their interactions can be approximated in physically based models. This thesis aimed to develop new physical understanding of precipitation-driven soil erosion processes on the laboratory flume scale. Experiments were carried out using the 6-m × 2-m EPFL erosion flume. In all experiments, different surface treatments were employed, with sediment concentration data collected from the flume effluent. The experimental data were analyzed using a mechanistic process-based soil erosion model, known as the Hairsine-Rose (H-R) model, with the goal of assessing the model's ability to describe the data. The H-R model simulates an arbitrary range of sediment size classes, a feature that is crucial to describe flume-scale soil erosion in detail. A range of related phenomena was investigated, as outlined below. First, the effect of raindrop splash and transversal width of the experimental flume on soil erosion was examined. In addition, the parameter consistency of the H-R model under these conditions was tested. Results showed that the H-R model represented well the total sediment concentrations as well as those of the fine and medium sediment size classes only when the transverse width of the flume experiment was below/around a threshold value related to lateral splash length. It was found, however, that boundary condition-induced asymmetry markedly reduces the applicability of the H-R model (and likely any other erosion model). Furthermore, in common with some existing reports, it was confirmed that the H-R model does not predict well the breakthrough of larger sediment size classes, possibly because for those sediments the transport mechanism involves rolling on the bed, which is not accounted for in the model. Second, the proportionality between soil erosion and area-exposed to the raindrop splash was tested. Results revealed that under carefully controlled laboratory conditions, the cumulative eroded mass depends on the cumulative runoff, and that soil erosion is proportional to the area exposed to raindrops. However, this relationship was controlled to a smaller extent by the initial soil conditions such as moisture content, bulk density and soil surface characteristics. The sediment concentrations in the flume effluent during the initial phase of the erosion event were sensitive to the initial state of the soil surface. At steady state, it was observed that the concentration of eroded sediments were controlled mainly by the effective rainfall and area exposed to raindrops. In contrast, using published data it was shown that rain splash erosion in the field is, in general, not proportional to the area exposed. Unlike the controlled laboratory experiments, the field experiments were characterized by a non-uniform initial surface roughness, surface soil aging and heterogeneous rock fragment size and spatial distribution. The third part of this study was dedicated to investigation of the effect of rock fragments on flume-scale hydrological response and soil erosion delivery. An area-based modification of the H-R model that accounted for the rock fragment coverage was employed to analyze the experimental data. The results revealed that the rock fragments protected the soils from raindrop splash and retarded the overland flow, therefore decreasing its sediment transport capacity. Individual rock fragments prevented surface sealing beneath them, which in turn increased the overall infiltration rate and affected the development of water depth compared with bare soil conditions. At short times, rock fragments were found to affect selectively the different size classes. This is due to the delayed (compared to a bare soil) development of the overland flow depth at the commencement of the erosion event. At steady state, the rock fragments led to decreased concentrations of the individual size classes in proportion to effective rainfall and the area exposed to raindrops. It was found that the modified H-R model predictions agreed well with measured sediment concentrations when high rainfall intensity and low rock fragment coverage were used. On the other hand, the H-R model could not reproduce the details of measured sediment breakthrough curves for conditions of low rainfall intensity and high rock fragment coverage. Finally, the effects of initial soil conditions and rock fragment coverage on particle size distribution and eroded mass were investigated using multiple rainfall events. Results revealed that short-time soil erosion is sensitive to the erosion history, in particular to whether steady-state conditions were reached in the preceding erosion event. Because the largest sediment sizes take longer to reach steady state, this sensitivity was more pronounced for the larger size sediment classes. The results also showed that the presence of rock fragments on the topsoil surface reduced the time needed to reach steady state compared with bare soil. This reflects the reduction in rain splash erosion caused by rapid development of the overland flow depth (which results from the reduction of the flow cross-sectional area by rock fragments).