A two-step framework to analyze local microstructure variations of paper sheets based on 3D image data is presented. First, a multi-stage workflow efficiently acquires a large set of highly resolved tomographic image data, which enables—in combination with statistical image analysis—the quantification of local variations and pairwise correlations of morphological microstructure characteristics on length scales ranging from micrometers to centimeters. Secondly, the microstructure is analyzed in terms of the local behavior of porosity, thickness, and further descriptors related to transportation paths. The power of the presented framework is demonstrated, showing that it allows one (i) to quantitatively reveal the difference in terms of local structural variations between a model paper before and after unidirectional compression via hard-nip calendering and that (ii) the field of view which is required to reliably compute the probability distributions of the considered local microstructure characteristics is at least 20 mm$^{2}$. The results elucidate structural differences related to local densification. In particular, it is shown how calendering transforms local variations in sheet thickness into marked local mass density variations. The obtained results are in line with experimental measurements of macroscopic properties (basis weight, Bekk smoothness parameters, thickness, and Gurley retention times).