To fuse transport vesicles with target membranes, proteins of the SNARE (soluble N-ethylmaleimide-sensitive factor attachment protein receptors) complex must be located on both the vesicle (v-SNARE) and the target membrane (t-SNARE). In yeast, four integral membrane proteins, Sed5, Bos1, Sec22 and Bet1 (refs 2-6), each probably contribute a single helix to form the SNARE complex that is needed for transport from endoplasmic reticulum to Golgi. This generates a four-helix bundle, which ultimately mediates the actual fusion event. Here we explore how the anchoring arrangement of the four helices affects their ability to mediate fusion. We reconstituted two populations of phospholipid bilayer vesicles, with the individual SNARE proteins distributed in all possible combinations between them. Of the eight non-redundant permutations of four subunits distributed over two vesicle populations, only one results in membrane fusion. Fusion only occurs when the v-SNARE Bet1 is on one membrane and the syntaxin heavy chain Sed5 and its two light chains, Bos1 and Sec22, are on the other membrane where they form a functional t-SNARE. Thus, each SNARE protein is topologically restricted by design to function either as a v-SNARE or as part of a t-SNARE complex.