28 pages, This study investigates how the emotional tone of food safety risk communication messages predicts message virality on social media. Through a professional Internet content tracking service, we gathered news articles written about the 2018 romaine lettuce recall published online between October 30th and November 29th, 2018. We retrieved the number of times each article was shared on Twitter and Pinterest, and the number of engagements (shares, likes, and comments) for each article on Facebook and Reddit. We randomly selected 10% of the articles (n = 377) and characterized the emotional tone of each article using machine learning, including emotional characteristics such as discrete emotions, emotional valence, arousal, and dominance. Conveying negative valence, low arousal, and high dominance, as well as anger and sadness emotions were associated with greater virality of articles on social media. Implications of these findings for risk communication in the age of social media are discussed.