Improvement of soil scattering prediction using radar remote sensing and electromagnetic modelling is partly related to precise and reliable multi-scale description of soil surface roughness. As the description of roughness-profiles is relatively time-consuming, there is a need for developing adequate sampling strategy. This work examines the within-field spatial variability of random and oriented surface roughness, respectively measured parallelly and perpendicularly to row crops, just after seeding (May 2003), in a corn field showing high pedodiversity. Surface roughness was recorded using a photograph of a grid board (0.6 m high x 1.0 m long, with 5 mm grid resolution). Parameters recorded were the root-mean-square (rms) height (a standard deviation), the correlation length (autocorrelation function) and the fractal dimension (Hausdorff-Besicovitch statistic). The field was stratified in three soil types (fine sandy loam, sandy clay loam and clay). Three sampling points were randomly selected per soil type. Two random and two oriented roughness-profiles were measured at each sampling point. Random roughness parameters show higher variability (CV) than oriented roughness. Soil types have a significant effect on random and oriented soil surface roughness parameters. An efficient sampling strategy is proposed for measuring soil surface roughness.