The numerical modelling of transport phenomena in porous media often requires a compromise between grid precision and the accuracy of simulation results. This study demonstrates the impact of errors on the accuracy of the reproduction of the actual pore space by the numerical grid on the estimated values of the saturated water conductivity. Four types of computational grids with varying levels of complexity were prepared for each of the 12 tomographic images of the porous specimens. The specific surfaces and total porosities were calculated for each of the meshes and compared with those parameters calculated for binarized tomographic images. Simulations of steady flow were performed on the computational grids, and the saturated water conductivity values were calculated. It has been shown that an insufficiently accurate mesh only reproduces the largest pore spaces in the analysed sample, which most often leads to an underestimation of the water conductivity coefficient. The following criterion for the optimal accuracy of the computational grid is proposed, it is based on the voxel size of the tomographic images of the porous media: the minimum size of the cell in the mesh used for simulations has to be at most two times the size of the voxel used in the tomographic scans of the porous medium.