This paper develops and tests a game-based model of enduring internati
onal rivalries. The model is embedded in a theoretical framework that
conceives of interstate conflicts as series of temporally related game
s. Transition across games is governed by learning, which occurs when
actors revise their perception of the opponent in response to previous
interactions. The model shows that learning can sometimes produce pat
terns of repeated conflict. An empirical analysis of four enduring riv
alries reveals a high incidence of conflictual games in their early ye
ars. As posited by the model, learning does account for game transform
ations (or evolutionary patterns), but changes in relative capabilitie
s also have an important effect on how actors define their preferences
and perceive the opponent.