In this paper, we re-examine the data from O'Neill's experiment involving a repeated, two-person, constant-sum game. We find that there is less evidence in support of the minimax hypothesis than indicated by O'Neill. There is strong evidence of serial correlation in players' choices, with several players displaying statistically significant dependence on the past moves of their opponents. We interpret this finding as evidence that the players themselves rejected minimax play as the appropriate model for their opponents' behavior. We find no evidence that players' behavior approached minimax behavior as players became more experienced.