 SWARM is a de novo amplicon clustering method that addresses two fundamental flaws and popular methods by iteratively clustering nearly identical amplicons using a local threshold and then refining results based on cluster internal structure and amplicon abundances. This approach produces robust operational taxonomic units in a fast, scalable, and input order independent manner, reducing the influence of clustering parameters. This article was authored by Frederic Mahé, Torbjorn Rohns, Christopher Quince, and others.