Very local spatial soil variability on Sudano-Sahelian coversands hampers the interpretation of agronomic research and is an obstacle for the dissemination of research findings. In an earlier paper, we specifically accounted for this spatial soil variability: Using novel tools for data exploration, such as non-parametric kernel density regression and spatial econometrics, on spatially explicit data for topsoil N, P, K and millet yield, we could explain 81% of the yield variation. However, the macronutrients explained a modest portion of millet yield only, while the good explanatory power was largely derived from spatial dependence/autocorrelation. This implies that variables other than N, P and K determine spatial variability of soils and millet growth. In the present paper we identify these variables and show that the proportions of the cations (Ca, Mg, K and Na) at the exchange complex, in combination with the Al saturation profile are the main source of spatial variability at this scale. Equations including these soil properties explain 82% of the millet yield variation, without spatial autocorrelation being present in the residuals. High proportions of Mg and Na in the topsoil coincide with low Al saturation levels in the subsoil and under such conditions yields are low (and vice versa). The likely operating mechanism is the destabilizing effect that Mg and Na have on the clay fraction, which, in turn, causes surface sealing. The latter affects seedling emergence and water infiltration and, consequently, subsoil Al saturation levels. The evidence provided further suggests a parent material connection to local soil variability: coversands of different source materials and age occurring as shallow layers. © 2003 Elsevier B.V. All rights reserved.