The differences in unsaturated permeability between concrete prepared in the laboratory (labcrete) and on a real construction site (fieldcrete) were investigated experimentally and using inverse analysis. Although more porous, the fieldcrete specimens were shown to have an average permeability very close to the labcrete ones. The fieldcrete permeabilities also exhibited more variability than the labcrete ones. Numerical simulations performed on 3D reconstructed concrete specimens showed that differences in coarse aggregate distribution cannot explain the variability observed at the macroscopic scale and that an important part may originate from the properties of mortar matrix.