We investigate the relationship between social capital and the decision to flee after a fatal road accident. This event is unplanned, and the decision is taken under great emotional distress and time pressure, thus providing a test of whether social capital matters for behaviour in extreme conditions. We merge data from the universe of fatality accidents involving pedestrians in the US over the period 2000–2018 with a unique dataset on social capital measures at the county level. Using within-state-year variation, our results show that one standard deviation increase in social capital is associated with a reduction in the probability of hit-and-run of around 10.5%. The causal interpretation of this evidence is supported by a number of falsification tests based on differences in social capital endowment between the county where the accident occurs and the county where the driver resides, as well as by the IV approach proposed by Lewbel (2012). Our findings show the importance of social capital in a new context, suggesting a broad impact on pro-social behaviour and adding to the positive returns of promoting civic norms.