Cervical cancer occurs significantly in women in developing countries every day and produces a high number of casualties, with a large economic and social cost. The World Health Organization, in the right against cervical cancer, promotes early detection screening programs by difeerent detection techniques such as conventional cytology (Pap), cytology liquid medium (CML), DNA test Human Papillomavirus (HPV), staining with dilute acetic acid and Lugol's iodine solution. Conventional cytology is the most used technique, being widely accepted, inexpensive, and with quality control mechanisms. The test has shown a sensitivity of 38% to 84% and a specificity of 90% in multiple studies and has been considered as the choice test for screening . The cervical cancer is not a public health problems in developed countries since more than three decades, among others because of implementation of other tests such as the CML which has increased the sensitivity to a figures that vary between 76% and 99 %. This test in particular produces a thin monolayer of cells that are examined. In our countries this technique is really far from being applied because of its high cost. In consequence, the conventional cytology has remained in practice as the only possible examination of the cervix pathology. In this technique, a sample of cells from the transformation zone of the cervix is taken, using a brush or wooden spatula, spread onto a slide and fixed with a preservative solution. This sample is then sent to a laboratory for staining and microscopic examination to determine whether cells are normal or not. This task requires time and expertise for the diagnosis. Attempting to alleviate the work burden from the number of examinations in clinical routine scenario, some researchers have proposed the development of computational tools to detect and classify the cells of the transformation cervix zone. In the present work the transformation zone is firstly characterized using color and texture descriptors defined in the MPEG-7 standard, and the tissue descriptors are used as the input to a bank of binary classifiers, obtaining a precision of 90% and a sensitivity of 83 %. Unlike traditional approaches that extract cell features from previously segmented cells, the present strategy is completely independent of the particular shape. Yet most works in the domain report higher precision rates, the images used in these works for training and evaluation are really different from what is obtained in the cytology laboratories in Colombia. Overall, most of these methods are applied to monolayer techniques and therefore the recognition rates are better from what we found in the present investigation. However, the main aim of the present work was thus to develop a strategy applicable to our real conditions as a pre-screening method, case in which the method should be robust to many random factors that contaminate the image capture. A segmentation strategy is very easily misleaded by all these factor so that our method should use characteristics independently of the segmentation quality, while the reading time is minimized, as well as the intra-observer variability, facilitating thereby real application of such screening tools.