In France, claystone geological formations are privileged candidates for deep underground nuclear waste storage, due, in particular, to their low permeability ability. However, in order to better predict the behavior of such media, the characterization of the pore network is essential. Such low permeability materials may have a high overall porosity but the majority of this porosity is constituted by pores smaller than 100 nm and whose distribution is still poorly known. The connectivity and the topology of these pores influence the porous media properties, for instance in regards to gas transport. In order to better understand the pore network structure, imaging techniques have been used to provide 2D and 3D images, such as microtomography or FIB-SEM. The latter allows a nanometric resolution (5 to 20nm) at the expense of the total imaged volume (usually a few µm3). Such a sample size is usually below the REV (Representative elementary volume), especially in regards to transport properties. In this project, we explore the possibility to distance the imaged 2D slices and use a numeric method to reconstitute the 3D volume at the REV scale. To this end, we study a stack of 180 parallel FIB-SEM slices of a synthetic clay of nanometric porosity. We compare two methods, based on multiple-point statistic algorithm (MPS), to reconstruct a coherent 3D configuration of its porous structure from 2D sections with different sampling spaces. The results of the reconstruction based on sparse data are then compared to the initial volume in regards to global characteristics – such as porosity - , morphological parameters – such as pore size distribution and connectivity index - as well as Lattice Boltzmann flow simulations to assess the capacity of the reconstruction method to provide satisfactory results. The first method, using a slice sequential reconstitution, produces results with low sensitivity to conditioning data but induces planar effects. The second method, aggregating values of three orthogonal directions allows a better reconstruction of the pore structure. This method is, however, more sensitive to conditioning data. The permeability values resulting from flow simulations on the second method are similar to those obtained on the total image for a distance between conditioning slices up to 11 cells. Thus, by reducing the number of direct images to acquire, this method opens the possibility of exploring volumes 5 to 10 times larger than those currently analyzed, therefore leading to a better representativity of the porous medium.