After playing the Chicken game against an anonymous random opponent, players report their beliefs about their opponent's action. It is argued that the reported beliefs are not consistent with the standard theory of rational choice, which predicts a negative correlation between a player's action in the Chicken game and the probability he assigns to his opponent choosing the same action. It is also argued that the reported beliefs are in uenced by self-similarity considerations, whereby a player tends to think that other players behave similarly to him, and thus reports beliefs that gravitate toward his own action. Self-similarity considerations are stronger when players are asked about the distribution of actions in the population of potential opponents than when they are asked about their particular opponent.