Soil sorption processes largely control the environmental fate of herbicides. Therefore accuracy of sorption parameters is crucial for accurate prediction of herbicide mobility in agricultural soils. A combined experimental and statistical study was carried out to investigate the small scale spatial variability of sorption parameters for atrazine and dinoseb in soils and to establish the number of samples needed to provide a value of the distribution coefficient (Kd) next to the mean, with a given precision. The study explored sorption properties of the two herbicides in subsurface samples collected from four pits distributed along a transect of an alluvial soil; two to four samples were taken at very short distances apart (about 30 cm), at each sampling location. When considering all the data available, the distribution coefficients were found to be normally and log-normally distributed for atrazine and dinoseb, respectively; the coefficients of variation were relatively high (close to 50% for dinoseb and 40% for atrazine). When analyzed horizon by horizon, the data revealed distribution coefficients normally distributed for both herbicides, whatever the soil layer, with lower coefficients of variation. The Kd values were shown to vary considerably between samples collected at very short distance (a few centimeters), suggesting that taking a single soil sample to determine sorption properties through batch experiments can lead to highly unrepresentative results and to poor sorption/mobility predictions.