This paper presents a new technique to prepare reduced-scale soil models with controlled spatial variability. Soil spatial variability is obtained analytically by using a discrete random field generation method. Heterogeneous models are then constructed by using nine homogeneous soils, which are created by mixing different percentages of kaolin and bentonite at water content of 1·5 times the liquid limit. Afterwards, the heterogeneous models are subjected to a consolidation process that is developed in two stages: oedometric path and geotechnical centrifuge. Finally, the mechanical soil properties of the heterogeneous models are estimated by measuring the unit weight and the water content of the random field elements. Results show that the mechanical properties depend not only on intrinsic behaviour, but also on space locations. The method proposed in this paper offers new possibilities for the study of the effect of soil variability on the behaviour of geotechnical structures.