Rapid neurotransmitter release depends on the ability to arrest the SNAP receptor (SNARE)-dependent exocytosis pathway at an intermediate "cocked" state, from which fusion can be triggered by Ca(2+). It is not clear whether this state includes assembly of synaptobrevin (the vesicle membrane SNARE) to the syntaxin-SNAP-25 (target membrane SNAREs) acceptor complex or whether the reaction is arrested upstream of that step. In this study, by a combination of in vitro biophysical measurements and time-resolved exocytosis measurements in adrenal chromaffin cells, we find that mutations of the N-terminal interaction layers of the SNARE bundle inhibit assembly in vitro and vesicle priming in vivo without detectable changes in triggering speed or fusion pore properties. In contrast, mutations in the last C-terminal layer decrease triggering speed and fusion pore duration. Between the two domains, we identify a region exquisitely sensitive to mutation, possibly constituting a switch. Our data are consistent with a model in which the N terminus of the SNARE complex assembles during vesicle priming, followed by Ca(2+)-triggered C-terminal assembly and membrane fusion.