An image matching or face pair matching is purely different aspect with respect to the other problems of computer vision and pattern recognition. This is a very active and challenging topic due to the unavailability of any prior information to the matching expert about the input images to be matched. Therefore an additional set of images can resolve this problem in some extent. In this context a cohort based face pair matching system is proposed. Initially the cohort set is common to all images but finally a subset of cohort images, specific to each of the paired images, are selected. Here Max-Min-Centroid-Cluster (MMCC) is applied which is capable enough to choose very relevant cohorts corresponding to target images. The raw similarity score between the input images is normalized with these set of cohort scores to obtain two normalized matching score. Afterwards the closeness between the images is measured by cross cohort normalization. The absolute difference of these two crossly normalized score is calculated and compared with a threshold value to decide the belonging of the input images to the same person or different person. The experiment has been conducted on ORL face database and the results found make evidence of the proposed system to be efficient.