The water-dissociation reaction (WD, H2O → H+ + OH−) affects the rates of electrocatalytic reactions and the performance of bipolar membranes (BPMs). How catalyzed interfacial WD is driven by voltage, however, is not understood. We designed a BPM electrolyzer with two reference electrodes attached laterally to each layer/side (here, poly(arylpiperidinium) and perfluorosulfonic-acid ionomers) to measure WD current and overpotential (ηwd), without soluble electrolyte and as a function of temperature and catalyst-layer properties. Using TiO2-P25 nanoparticles as a model WD catalyst, Arrhenius-type analysis yields a WD activation energy Ea of 25–30 kJ mol−1, only weakly dependent ηwd. The pre-exponential factor is unexpectedly proportional to ηwd. With D2O, ηwd is ~2 to 4 times larger than in H2O, largely due to a lower pre-exponential factor. Without catalyst, ηwd is ~10-fold larger and Ea decreases from 34 to 24 kJ mol−1 as ηwd goes from 0.1 to 1 V. To explain these data, we propose a new WD mechanism where metal-oxide nanoparticles, polarized by the voltage across the BPM junction, serve as i) proton acceptors (from water) on the negative sides of the particle to generate free OH−, ii) proton donors on the positive sides to generate H3O+, and iii) surface proton conductors that connect spatially separate donor/acceptor sites. Increasing electric-field strength with overpotential orients water for proton-transfer elementary steps comprising WD, increasing the pre-exponential factor and hence rate, but is insufficient to lower Ea. This understanding will accelerate development of electrocatalysis, electrodialysis, carbon-capture, and carbon-utilization technologies that require efficient WD.