Members of the serpin (serine protease inhibitor) superfamily fold into a metastable conformation that is crucial for proper function. As a consequence, serpins are susceptible to mutations that cause misfolding and the intracellular accumulation of pathogenic polymers. The mechanism of serpin polymerisation remains to be resolved, however, over the past two decades the loop-sheet hypothesis has gained wide acceptance. In this mechanism the reactive centre loop of one serpin monomer inserts into the beta-sheet A of another (in trans), in a manner similar to what is seen for reactive centre loop-cleaved and latent conformations (in cis). The hypothesis has been refined in response to certain experimental data, but it has proved difficult to assess the various propositions without creating molecular models. Here we evaluate the loop-sheet mechanism by creating models of pentamers of the archetypal serpin alpha(1)-antitrypsin. We conclude that an inescapable consequence of the loop-sheet mechanism is polymer compaction and rigidity, properties that are inconsistent with the beads-on-a-string morphology of polymers obtained from human tissue. The recent crystal structure of a domain-swapped serpin dimer suggests an alternative mechanism that is consistent with known polymer properties, including the requirement of partial unfolding to induce polymer formation in vitro, and polymerisation from a folding intermediate in vivo.