Sediment is fractionated by size during its cascade from source to sink in sediment routing systems. It is anticipated, therefore, that the grain size distribution of sediment will undergo down-system changes as a result of fluvial sorting processes and selective deposition. We assess this hypothesis by comparing grain size statistical properties of samples from within the erosional source region with those that have undergone different amounts of transport. A truncated Pareto distribution describes well the coarser half of the clast size distribution of regolith, coarse channel bed sediment and proximal debris flows (particularly their levees), as well as the coarser half of the clast size distribution of gravels that have undergone considerable amounts of transport in rivers. The Pareto shape parameter a evolves in response to mobilization, sediment transport and, importantly, the selective extraction of particles from the surface flow to build underlying stratigraphy. A goodness of fit statistic, the Kolmogorov–Smirnov vertical difference, illustrates the closeness of the observed clast size distributions to the Pareto, Weibull and log-normal models as a function of distance from the depositional apex. The goodness of fit of the particle size distribution of regolith varies with bedrock geology. Bedload sediment at catchment outlets is fitted well by the log-normal and truncated Pareto models, whereas the exponential Weibull model provides a less good fit. In the Eocene Escanilla palaeo-sediment routing system of the south-central Pyrenees, the log-normal and truncated Pareto models provide excellent fits for distances of up to 80 km from the depositional apex, whereas the Weibull fit progressively worsens with increasing transport distance. A similar trend is found in the Miocene–Pliocene gravels of the Nebraskan Great Plains over a distance of >300 km. Despite the large fractionation in mean grain size and gravel percentage from source region to depositional sink, particle size distributions therefore appear to maintain log-normality over a wide range of transport distance. Use of statistical models enables down-system fractionation of sediment released from source regions to be better understood and predicted and is a potentially valuable tool in source-to-sink approaches to basin analysis.