THE ECONOMIC LITERATURE concerning agents' reputations has grown steadily since the seminal work of Kreps, Milgrom, Roberts, and Wilson.2 Early work focused on how incomplete information leads to equilibria that are vastly different (but more intuitive) than those possible in the complete information game. Recently, however, game theorists have been studying how incomplete information might refine the set of equilibria.3 One important class of games is that in which a single long-run agent plays a simultaneous move (stage) game with a sequence of opponents, each of whom plays only once, yet observes all previous play. Fudenberg and Levine (1989) study the reputation of the long-run player in this type of game. They argue that the 'most reasonable' equilibrium is the one which the long-run player most prefers. Their intuition is sustained when one perturbs the game with the Stackelberg strategy. Fudenberg and Levine show that in the perturbed game the equilibrium payoffs of the long-run player are bounded below by a number that converges to the Stackelberg payoff.