Soils are important sources of sediment and phosphorus in rural catchments, necessitating the development of mathematical models for impact assessment. In this paper, multiple empirical models are tested on an event basis at four nested locations in an intensively managed grassland headwater catchment while accounting for parameter and data uncertainties using extended Generalized Likelihood Uncertainty Estimation (GLUE). The study provides the first template of model comparison under data uncertainty in soil erosion and phosphorus transfer modelling as well as hypotheses of soil and water processes in the study catchment. A fodder field, yielding large sediment and phosphorus concentrations in runoff, is characterized by inter-event variation in sediment-discharge relationship, mild intra-event hysteretic behaviour and seemingly random erosion incidents. Sediment-discharge variation is partly formalized by parameter variation as a function of antecedent soil moisture, indicative of a gradual shift from transport- to source-limited behaviour, decreasing soil erodibility and/or decreasing initial flow erosivity and transport capacity with increasing antecedent wetness. The catchment outlet appears to be source-limited while converging flows with different sediment concentrations, variable erosion processes and/or sporadic entrainment of near- or in-stream sediments gain importance. Phosphorus dynamics are strongly linked to those of sediment. Non-linearities can be explained by preferential transfer of phosphorus-rich organic matter at small flows while there is no significant evidence of preferential transfer of phosphorus-rich mineral fines. Iterating between collecting data, constraining uncertainties and rejecting and improving models is suggested as a consistent framework for understanding soil erosion and phosphorus movement.