Changing the scale of observation or averaging has a significant, but poorly understood, impact on the apparent variability of hydrological quantities. The representative elementary area (REA) concept is used as a motivation for measuring inter-storm streamflow and calculating wetness index distributions for the subcatchments of two small study areas in New Zealand. Small subcatchments are combined to provide larger scale samples, and then the variance of specific discharge between similar sized subcatchments is calculated. For small subcatchments (area less than similar to 1 km(2)) this variance is found to decrease with area more quickly than might be expected if the catchments were random samples. Such behaviour is tentatively interpreted as evidence supporting the concept of 'organization'. At larger scales, variance between catchments decreases in a way that is consistent with sampling from a stationary random field. The results from the streamflow data are reinforced by an analysis of topographic data for the two study areas, although some questions remain open.Both the flow and topographic data support the idea that it is possible to find an averaging scale where the variability between catchments is sufficiently small for a 'distribution function' approach to be used in distributed rainfall-runoff modelling. Consistent estimates of the scale at which the study areas become stationary (0.5 km(2) for Little Akaloa, 2 km(2) for Lewis) are obtained using both flow and topographic data. The data support a pragmatic REA concept which allows meaningful averages to be formed: this may be a useful base for further conceptual developments, but it is not appropriate for a classical continuum approach. Further conceptual development combined with field measurement and computer simulation are still required for the REA to have operational impacts. In particular, it is not clear which models are appropriate for use at the REA scale.