Flocking birds frequently look up or "scan" while they are feeding on the ground. High scanning rates increase the probability that the birds in a flock will see an approaching predator in time to avoid predation; however, high scanning rates also decrease the feeding rates of the scanning individuals. Since the scanning rate that maximizes the survival probability of one individual depends on how frequently other birds in the same flock are scanning, the optimal scanning behavior must be modeled as a game. We develop a realistic model of scanning behavior and use it to find two game theoretical solutions-the "co-operative" or Pareto optimum and the "selfish" or Nash optimum. The observed scanning rates do not differ significantly from the co-operative optimal scanning rate. We argue that in a game where players meet again and again such apparent co-operation may be an evolutionarily stable strategy.