People often engage in protest because others do so as well, not because their grievances or political conditions have fundamentally changed. Consequently, mass movements can erupt suddenly. Yet, the explosive character of many popular upheavals constitutes an unsolved puzzle in the study of protest participation. This paper tackles this puzzle by using agent-based models. It maps out the decisive role of critical thresholds - chain reactions by which small numbers of people can trigger mass participation. This paper’s main innovative feature is that simulation outcomes are confronted with empirical data: validation is an important direction in which computational modelling approaches need to move. Data on the occurrence of demonstrations are drawn from the domestic conflict data of the Cross-National Time-Series Data Archive (Banks & Wilson 2013). To assess the impact of protests on regime changes, a recent data set on autocratic regimes (Geddes, Wright & Frantz 2013) is used.