**Digital Traces of Offline Mobilization** Since 2009, there has been an increase in global protests and related online activity. Yet, it is unclear how and why online activity is related to the mobilization of offline collective action. Across two analyses, we used digital traces of online behavior and data science techniques to model people’s online and offline behavior around a mass protest. In Study 1a, we used Twitter behavior posted on the day of the protest by attendees or non-attendees (759 users; 7,592 tweets) to train and test a classifier that predicted, with 80% accuracy, who participated in offline collective action. Attendees tended to be co-present online and offline, and used their mobile devices to broadcast their offline presence and plan logistics. In Study 1b, using the longitudinal Twitter data and metadata of a subset of users from Study 1a (209 users; 277,556 tweets), we found that participation in the protest was not associated with an individual’s online polarization over the year prior to the protest, but it was positively associated with the validation (“likes”) they received on their relevant posts. These studies provide an empirical demonstration of the features of online interactions that are related to offline action; new methodological approaches that connect online behavior with offline collective action, including a novel equation that indexes online polarization, and two algorithms that could inform crowd management and emergency responses. Using these new methods, we explain how studying online behavior enables novel conceptual insights into how and why people mobilize online, to take offline collective action.