Recently we demonstrated that the unique stereoelectronic relationships inherent in the structure of plasmenylethanolamine facilitate membrane fusion, and we postulated the existence of a membrane fusion protein which could exploit the propensity of plasmenylethanolamine molecular species to adapt an inverted hexagonal phase [Glaser & Gross (1994) Biochemistry 33, 5805-5812], We now report a cryptic membrane fusion activity in rabbit brain cytosol, which requires separation from an endogenous inhibitor to express its activity, and demonstrate that vesicle fusion catalyzed by this protein is highly selective for membrane vesicles containing plasmenylethanolamine. The cytosolic protein catalyzing membrane fusion activity was purified to apparent homogeneity by sequential column chromatographies, revealing a single 38-kDa protein band after sodium dodecyl sulfate-polyacrylamide gel electrophoresis and silver staining. Automated Edman degradation demonstrated that the purified protein is an isoform of glyceraldehyde-3-phosphate dehydrogenase (GAPDH), which was confirmed by Western blot analysis utilizing polyclonal antibodies and by solution-state inactivation of membrane fusion activity by a monoclonal antibody directed against GAPDH. Both GTP-affinity and Mono Q chromatographies resolved GAPDH isoforms that catalyzed dehydrogenase activity from the GAPDH isoform that catalyzed membrane fusion activity. The purified fusion protein was calcium-independent, resistant to treatment with N-ethylmaleimide, and possessed an obligatory requirement for plasmenylethanolamine and cholesterol. High-resolution stopped-flow kinetic analysis of plasmenylethanolamine-facilitated membrane fusion demonstrated that one tetramer of the GAPDH isoform catalyzed one fusion event between two vesicles containing plasmenylethanolamine every millisecond (on average). Collectively, these results constitute the first description of a protein which can catalyze the fusion of vesicles at a rate which satisfies the mathematical constraints imposed by the observed rates of fusion of synaptic vesicles with the presynaptic membrane in vivo.