Rivalry in the Vancouver retail gasoline market is modeled as a repeated game. Service-station demand, cost, and reaction functions are estimated from daily data on individual station prices, costs, and sales. These functions are then used to calculate noncooperative and cooperative solutions to the constituent game and the actual outcome of the repeated game. The actual outcome is found to be substantially less lucrative than the monopoly solution. Nevertheless, all stations are better off than if they played their noncooperative strategies in every period. In addition, the continuous supergame strategies associated with reaction functions are found to provide a better model of the price-war dynamics than alternative discontinuous strategies, where price wars are reversions to Nash behavior.