We have developed an automated system dedicated to computer aided diagnosis for serous effusion cytology. The system can take as inputs, images from different slide scanners or microscope (ScanScope CS Aperio Technologies, x20, or SCN400 Leica Microsystems, x40, or Olympus BX 50 microscope equipped with a Märzhäuser motorized stage x20). We have considered 150 positive and 50 negative cases of serous cytology stained with the Papanicolaou standard. The system considers each virtual slide as a set of tiled images. Each tile is processed in three steps: image segmentation to extract cytoplasm and nuclei, cell feature extraction (50 features ranging from shape to texture), cell classification (in the following categories: polynuclear, lymphocyte, macrophage, normal and dystrophic mesotheliale cells, abnormal cells). Image segmentation relies on automatic pixel classification and mathematical morphology. This segmentation has been evaluated on more than 10.000 cells with 96 % of correct segmentations. Cell classification relies on a dedicated architecture of neural networks. The performance of this architecture has been evaluated on 150 cases that were diagnosed by pathologists as positive and containing various metastatic adenocarcinoma, mesothelioma, and atypical hyperplasic mésothèliale cells. The system has been able to recognize 98.5% of these cases. On 50 cases diagnosed as negative, the system has been able to identify 4 cases as suspect. For these cases, the amount of suspect cells was around 0.3% of the total number of cells. Pathologists have reviewed these cells in a second reading, and agreed with classifying these cases as suspicious. This validation study has therefore enabled to show that the conceived system is a valuable aid in the detection of cancer cells for early diagnosis of tumors.