In the United States many antidumping petitions are withdrawn before the investigations are completed. Prusa (1992) argues that petitions are used by domestic industries to induce foreign industries into collusive agreements. In his model, all antidumping petitions should be withdrawn, which is not the case. This paper provides a model in which only some petitions are withdrawn. Withdrawal depends on two key parameters: coordination cost and bargaining power of domestic and foreign industries. A new data set is constructed to test the model on the U.S. experience for the period 1980–97. The econometric analysis supports the theoretical conclusions of the model.