The manual signature-based authentication of a large number of documents is a laborious and time-consuming task. Consequently many off-line signature verification sys- tems were recently developed. In this paper we propose a human-centric system, which exploits the synergy between human and machine capabilities, and show that this com- bined system can perform better (than humans or a ma- chine) for almost all operating costs. The combination stra- tegy is based on techniques in receiver operating charac- teristics (ROC) analysis. We conduct an experiment on a data set that contains 765 test signatures from 51 writers, and record the performance of 23 human classifiers, and that of a hidden Markov model-based (HMM-based) clas- sifier, in ROC space. We propose that a manager (human or machine) specifies acceptable operating costs (Neyman- Pearson criterion), after which our human-centric system makes an optimal decision by utilizing the maximum attain- able combined classifier.