Overland flow and soil erosion models working with coarse to medium resolution input data for prediction on large areas often rely on high resolution calibration and evaluation datasets collected for smaller areas. Therefore, one important issue is to assess to what extent process identification and parameterisation carried at one scale can be extrapolated to a different scale. The objective of this paper is to identify the nature and the cause of variation in the runoff response for areas ranging from plots to 10 km(2) catchments on the basis of extensive multi-scale field information gathered on arable lands of the loam belt in Northern France. Three datasets are studied: data from plots of ca. 500 m(2) consisting of either 100 or 90% arable land; data from a catchment of ca. 90 ha with either 96 or 94% arable land; and data from a catchment of ca. 1100 ha with ca. 80% arable land. For the three datasets mean slope gradient and soil types are similar. Between the three datasets, a significant decrease in the runoff coefficient is observed as the area increases. This trend, which has already been observed by many researchers, confirms that a catchment cannot be considered as the sum of individual fields. A simple distributed approach, based on the dominant processes, that defines the infiltration parameters at the field scale (allowing reinfiltration) and accounts for natural or human induced preferential pathways within the catchments could reproduce the observed runoff coefficients. In this context, the scaling transition between the plots and the 10 km(2) catchment can thus be considered in terms of connectivity between the 'infiltrating' and 'runoff producing' areas. However, if we select data with contrasting land use for the three datasets, it shows that the variations between the different runoff coefficients can be smoothed out or accentuated. Furthermore it demonstrates the importance of possessing a database that covers a large range of situations when studying spatial scaling linkages in order to be able to distinguish between the different factors. Finally these results are compared with results from a study where rainfall-runoff events were simulated on 40 selected catchments of ca. 10 km(2) in Hte-Normandie (France). For catchments of this scale, it appears that the percentage of arable land is a driving factor in the runoff response whatever its location within the catchment. One explanation being that connectivity variability can be significant between the plot scale and the catchment scale but is relatively less important between different 10 km(2) cultivated catchments. (C) 2004 Elsevier B.V. All rights reserved.