In a retrospective investigation for a new image-analytical nuclear grading method, we used 145 routine hematoxylin and eosin-stained, paraffin-embedded tissue sections from node-negative breast carcinomas. Cell fields of primary tumors were scanned in a light microscope in successive focus levels in 1-micron steps for thick sections (> or = 5 microns: quasi-3D analysis) and in one focus position for thin sections (< 5 microns: 2D analysis). After image-segmentation, nuclear features for texture and chromatin distribution were calculated. A binary classification tree was constructed for determination of two mathematically defined classes of high- and low-risk tumor cell nuclei. After fixing a cut-point for the portion of high-risk tumor cell nuclei per patient, it was possible to distinguish two different groups with significantly different relapse rates of 4.2% and 74.5% in quasi-3D analysis and 0.0% and 52.0% in 2D analysis, respectively. Large differences between quasi-3D and 2D analysis were only present in the classification of nonrelapse patients, whereas nearly all patients with relapse had more than 50% high-risk tumor cell nuclei. The results show that the information in thicker tissue sections contains important additive components in the third dimension, with respect to the detection of chromatin structure and distribution. This advantage should be exploited for the development of an objective image-analytical nuclear grading system as a highly significant prognostic marker.