This paper investigates feature transfer in classifier-trained networks, analyzing the impact of similarity between training and unseen data on unseen data clustering performance and feature extraction.