Résumé : In the present study we used computer-assisted microscopy to analyze the morphology of Feulgen-stained cell nuclei in cell populations obtained at the same time as routinely performed cervical smears and in the same way. We investigated in a series of 110 cases whether the quantitative morphonuclear description of cytological cervical samples is able to aid pathologists to distinguish between benign and more suspect premalignant lesions. For this task nuclear DNA content, nuclear morphometry (size and anisonucleosis level) and chromatin pattern-related parameters were compiled for each specimen enrolled in the database. A set of 32 normal and 17 high-grade squamous intraepithelial lesion (HSIL) specimens (with diagnostic confirmations) were selected as references and used to establish a discriminant model on the basis of cytometry-generated variables. This model was then used to score the remaining 61 cases in our series (including cases exhibiting benign cellular changes, squamous cells of undetermined significance, low-grade SIL and cancers). The results show that a model discriminating efficiently between normal and HSIL groups can be obtained by combining 5 quantitative features (1 DNA ploidy-related, 2 morphometrical and 2 chromatin texture features). A 97% specificity and an 88% sensitivity characterized the boundary so established. When applied to new cases, the model was in fact able to correct diagnoses for cases which had been down- or up-graded on the basis of the Bethesda system, and provided scores in accordance with histological control.