Despite its relevance to sensory features and to fundamental explanation of the changes observed throughout cheese ripening, microstructural studies of specialty cheeses have lagged far behind those of industrialized cheeses. Hence, the purpose of this study was to pinpoint microstructural differences in the gel network of traditional Serra da Estrela cheese throughout ripening, using 2-dimensional image analysis, and to unfold correlations of such microstructural indicators with classical bulk chemical and textural parameters. Hence, samples were taken throughout the ripening period, following a nested design; uniform thin sections were systematically observed via light microscopy (LM, 200 x) and transmission electron microscopy (TEM, 4,400 x), and computer-assisted quantitative analysis of digital images was comprehensively performed following standard stereological methodology. Fresh cheeses exhibited the highest porosity and ratio of surface area to volume. Significant negative correlations were found between microstructural parameters and proteolysis indicators. Light microscopy images suggested that rearrangements exist, up to 21 d, of the cheese matrix that leave porosity and pores unchanged, whereas TEM images indicated a significant decrease in number of pores within the same time frame, especially those above 1 x 10(-2) mu m(2) in area. The larger pores, chiefly with cross-sectional areas above 40 mu m(2), were less represented by the end of ripening-and likely explain the observed significant decrease of cheese porosity without a change in number of pores. Field viewing significantly affected the microstructural parameter values, whereas section viewing affected significantly only LM-based ones. Categorical principal component analysis between the 2 types of microstructural data sets was performed, and permitted discrimination of each stage of ripening. Multiple linear regression analysis indicated that the variables associated with the nitrogen fraction were well predicted by stereological-based parameters (R-2 = 0.96). Therefore, our findings demonstrate the potential of image analysis to monitor microstructure throughout ripening, and that the microstructure revealed by LM reflects more clearly cheese aging than that revealed by TEM.