The automatic analysis of cytological preparations is very challenging due to complexity of cellular structures. Classical segmentation methods such as thresholding, active contours or watershed are effective only for simple cases where nuclei are well isolated from each other. Unfortunately, in case of cytological material this requirement is very rarely fulfilled. To tackle this problem, I propose the method which transforms cytological image into simpler form where nuclei are approximated by circular or elliptical-shaped objects. For this purpose, it is assumed that cytological sample formation process can be described by marked point process. Thanks to this, it is possible to estimate the configuration of objects consistent with the observed image using maximum a posterior (MAP) estimation. The accuracy of this approach is evaluated with the help of referenced, manually segmented images. Degree of matching between reference nuclei and discovered objects is measured with the help of Jaccard distance and Hausdorff distance. Finally the accuracy of proposed approach is compared with the accuracy of Circular Hough Transform (CHT). The test set contains 50 cytological images, fragments of virtual slides of breast cancer tissue. Obtained results are promising because the method outperforms CHT in false positive (FP) rate. The model of cytological image can be used to seeds other, more precise segmentation methods, analysis of morphometric features or analysis of spatial topology of the material in the sample.