Soil moisture is a significant variable in its importance to the validation of hydrological models, but it is also the one defining variable that ties in all components of the surface energy balance and as such is of major importance to climate models and their surface schemes. Changing the scale of representation (e.g. from the observation to modelling scale) can further complicate the description of the spatial variability in any hydrological system. We examine this issue using soil moisture and vegetation cover data collected at two contrasting spatial scales and at three different times in the snow‐free season from a cutover peat bog in Cacouna, Québec. Soil moisture was measured using Time Domain Reflectometry (TDR) over 90 000 m2 and 1200 m2 grids, at intervals of 30 and 2 m respectively. Analyses of statistical structure, variance and spatial autocorrelation were conducted on the soil moisture data at different sampling resolutions and over different grid sizes to determine the optimal spatial scale and sampling density at which these data should be represented. Increasing the scale of interest without adequate resolution in the measurement can lead to significant inconsistency in the representation of these variables. Furthermore, a lack of understanding of the nature of the variability of soil moisture at different scales may produce spurious representation in a modelling context. The analysis suggests that in terms of the distribution of soil moisture, the extent of sampling within a grid is not as significant as the density, or spacing, of the measurements. Both the scale and resolution of the sampling scheme have an impact on the mean of the distribution. Only approximately 60% of the spatial pattern in soil moisture of both the large and small grid is persistent over time, suggesting that the pattern of moisture differs for wetting and drying cycles. Copyright © 2003 John Wiley & Sons, Ltd.