In this paper we construct a measure of adult psychological resilience, document its distribution in the population, and test its predictability by childhood socioeconomic circumstances. To do this we use a dynamic finite mixture model applied to 17 years of nationally representative panel data, and focus on the reaction to ten major adverse life events, including serious illness, major financial events, redundancy and crime victimisation. Importantly, our model accounts for non-random selection into events, anticipation of events, and differences between individuals in the immediate psychological response and the speed of adaptation. We find considerable heterogeneity in the response to adverse events, with the total psychological loss of individuals with low resilience being several times larger than the average loss. We also find that resilience is strongly correlated with clinical measures of mental health, but only weakly correlated with measures of cognition and personality. Finally, we find that resilience in adulthood is predictable by childhood socioeconomic circumstances; the strongest predictor we identify is good childhood health.