Selection of cohort models plays a vital role to increase the accuracy of a biometric authentication system as well as to reduce the computational cost. This paper proposes a novel approach for cohort selection called Max-Min-Centroid-Cluster (MMCC) method. The clusters of cohorts are generated by K-means clustering technique. The union of the clusters having largest and smallest centroid value is taken as cohort subset. The cohort scores, after normalization using different cohort based score normalization techniques, are used in authentication process of the system. Evaluation has been carried out on FEI face datasets. The performance of this novel methodology is analyzed using T-norm and Aggarwal (max rule) normalization techniques. Experimental results exhibit the efficacy of the proposed method.