Biometric Identification Systems capture biometric data i.e. face, fingerprint, iris, etc. images and store them in the database to form a Gallery. During an identification search, incoming biometric probe data is matched against all the images in the Gallery. A decision threshold is applied to these matches to obtain the potential $match(es)$ in the Gallery for a particular probe. Typically the above approach works very well if the gallery size is up to few thousands. However, performance degrades rapidly when the gallery size grows to few hundred thousands or millions of biometric records. We present an approach to drastically reduce the performance degradation without degrading the accuracy of the overall system.