Widely used models of the soil erosion process, such as the empirical universal soil loss equation (USLE) and revised USLE (RUSLE), suggest that soil loss, i.e. the mass of soil lost by water erosion per unit surface area, increases with slope length, λ, although contrasting experimental results may be found in the literature. Experiments were carried out at the Sparacia site, Sicily, to establish the influence of plot length on soil loss. This was tested using data characterised by a high number of replicated, bare plots of different lengths (0.25, 0.4, 1, 2, 5, 11, 22, 33 and 44 m), simultaneously operating in the 1999–2008 period. Event soil losses did not vary significantly with λ or even decreased as λ increased. A possible reason, which requires additional investigation, is that increasing λ had a moderate increasing effect on the rate of rill erosion but had an appreciable decreasing effect on the rate of interrill erosion. It was concluded that for the sampled site shortening the slope length is not an effective method to control total (i.e. rill + interrill) soil loss. The relative variability of the plot data decreased as the mean measured variables (soil loss, runoff, sediment concentration) increased. Relationships for estimating the optimum number of replicates for erosion studies were proposed.