Parochial altruism suggests that humans are intrinsically motivated to harm out-groups, and that this is tightly connected to a preference for benefitting their in-group. Yet, there is little evidence for the kind of unconditional out-group harm suggested by this account, nor for the assertion that it would be associated with in-group cooperation. Instead, humans selectively reciprocate actual, but also potential aggression. We therefore posit a model of parochial reciprocity, according to which individuals retaliate against actual and anticipated harms to their in-group. To test predictions arising from these competing accounts, we manipulated out-group threats and elicited preferences for the welfare of in-group and out-group members, as well as beliefs about in-group and out-group members' behaviours in an incentivised intergroup conflict game with natural groups (online sample; N = 973). In this game, individuals could pay to benefit their in-group, but had the option to additionally harm the out-group without incurring any further costs. Individuals who valued their in-group more strongly were no more likely to harm the out-group, thus contradicting parochial altruism. Instead, individuals who expected the out-group to harm their in-group preemptively retaliated the anticipated attack. Importantly, they only did so when the out-group posed an actual threat to the in-group. Taken together, the findings suggest that participation in intergroup conflict is better explained by parochial reciprocity than purely by group-based preferences.