We show that armed conflict affects social capital as measured by trust and associational membership. Using the case of Uganda and two rounds of nationally representative individual-level data bracketing a large number of battle events, we find that self-reported generalized trust and associational membership decreased during the conflict in districts in which battle events took place. Exploiting the different timing of two distinct waves of violence, we provide suggestive evidence for a rapid recovery of social capital. Evidence from a variety of identification strategies, including difference-in- difference and instrumental variable estimates, suggests that these relationships are causal.