For low power fuel cells, management of reactants, water and heat, must be realized in a passive fashion in order to minimize parasitic losses. Effective fuel, oxygen supply and water management for reliable performance are also greatly affected by cell geometry and materials. These are complex systems to optimize on a mere experimental basis. As an aid to this goal, data-driven analysis techniques, requiring no a priori mathematical model, are gaining a reputation in other research fields, where phenomenological modeling approaches might be intractable. In this work a 12 W PEMFC series stack and its individual cells are characterized by means of a new data-driven technique, M-NMF, combined with an alternating least squares algorithm. The PEMFC was operated for 12 combinations of air to H2 flow rate ratios, generating as many polarization curves. These data were grouped and differentiated according to similar over-potential patterns and insight gained into their relative contribution to stack performance. © 2010 Published by Elsevier Ltd on behalf of Professor T. Nejat Veziroglu.