Current models for Word Sense Disambiguation (WSD) struggle to disambiguate\nrare senses, despite reaching human performance on global WSD metrics. This\nstems from a lack of data for both modeling and evaluating rare senses in\nexisting WSD datasets. In this paper, we introduce FEWS (Few-shot Examples of\nWord Senses), a new low-shot WSD dataset automatically extracted from example\nsentences in Wiktionary. FEWS has high sense coverage across different natural\nlanguage domains and provides: (1) a large training set that covers many more\nsenses than previous datasets and (2) a comprehensive evaluation set containing\nfew- and zero-shot examples of a wide variety of senses. We establish baselines\non FEWS with knowledge-based and neural WSD approaches and present transfer\nlearning experiments demonstrating that models additionally trained with FEWS\nbetter capture rare senses in existing WSD datasets. Finally, we find humans\noutperform the best baseline models on FEWS, indicating that FEWS will support\nsignificant future work on low-shot WSD.\n