Screening is a manual activity which involves its subjectivity. A semi-automated computer-based system could contribute to the detection of screening errors by the way of a greater reliability. We intend to design such a system operating on color images from serous cytology. The aim of this paper is to present the strategy of the first part of the system: segmentation. It is based on mathematical morphology tools such as watersheds using color infom1ation in several color spaces. An extension of watershed to an optimal region-growing operator has been used. A pool of cells has been evaluated by experts to score the segmentation success rate. All cells have been isolated whatever their spatial configuration may be. The average success rate is 94.5% for the nuclei and 93% for the cytoplasm. Our morphological color segmentation of cytological serous images is accurate and provides a good tool for the extraction of cells.