Karyotyping is a set of procedures, in the scope of the cytogenetics, that produces a visual representation of the 46 chromosomes observed during the metaphase step of the cellular division, called mitosis, paired and arranged in decreasing order of size. Automatic pairing of bone marrow cells is a difficult task because these chromosomes appear distorted, overlapped, and their images are usually blurred with undefined edges and low level of detail. In this paper, a new metric is proposed to compare this type of chromosome images toward the design of an automatic pairing algorithm for leukemia diagnostic purposes. Besides the features used in the traditional karyotyping procedures, a new feature, based on mutual information , is proposed to increase the discriminate power of the G-banding pattern dissimilarity between chromosomes and improve the performance of the classifier. The pairing algorithm is formulated as a combinatorial optimization problem where the distances between homologous chromosomes are minimized and the distances between nonhomologous ones are maximized. The optimization task is solved by using an integer programming approach. A new bone marrow chromosome dataset-Lisbon-K1 (LK1) chromosome dataset with 9200 chromosomes---was build for this study. These chromosomes have much lower quality than the classic Copenhagen, Edinburgh, and Philadelphia datasets, and its classification and pairing is therefore more difficult. Experiments using real images from the LK1 and Grisan et al. datasets based on a leave-one-out cross-validation strategy are performed to test and validate the pairing algorithm.