To date, circulating tumor cells (CTCs) are the most promising tumor marker. They correlate with overall survival rate and disease free survival, allowing an early detection of metastatic process, monitoring the disease progression and the treatment response. Different from most state of the art methods that detect CTC's in stained blood, the aim of this paper is to identify CTCs in unstained blood in order to accomplish the conditions for long term monitoring. Thus, our approach is to find the best features that characterize CTCs and discriminate them from other blood cells in dark field microscopic images. Several classic texture features, such as histogram statistics, gray level co-occurrence matrix and gray tone difference matrix, were proposed as cell descriptors. In addition, we introduce new features that quantify the radial homogeneity of the cells. The study was performed for three types of cells: red cells, white cells and CTCs. The images in our study were acquired with a microscope in dark field (DF) mode at 10X and 20X optical magnification. Several classifier were designed based on the computed features. The performance of each type of feature was tested, and ranked. Final classification results are given by a simplified set of features that improve the quality of the classifiers. The accuracy of our results is over 98% for the cell classification in both optical magnifications.