In recent years several countries have made major efforts to coordinate trade and industrial policies. Examples include the United States-Canada Free Trade Agreement and the European attempt to complete the E.C. internal market by the year I992. The design of welfare-improving policy reform is complicated by the presence of a large number of policy-induced distortions and by the imperfectly competitive nature of many industries. It is well known that the nominal reduction of a set of distortions, holding other distortions fixed, does not necessarily improve welfare. This conclusion, which arises because the reform occurs in a second-best environment, certainly survives the introduction of imperfect competition. We examine additional complications that arise when reform is carried out under conditions of imperfect competition. This subject, which is of general theoretical interest, is also of immediate practical interest due to the current attempts to harmonise European trade and industrial policies. The principal conclusion is that limited cooperation, as occurs in customs unions, may be worse for members of the union than no cooperation. This possibility arises because comparison of welfare under the two situations (no cooperation and partial cooperation) amounts to a comparison of two second- best equilibria. Such a comparison is, in general, ambiguous. An alternative explanation of the result is that different degrees of cooperation induce different games. A comparison of the equilibria of these different games is, in general, ambiguous. In view of this, the possibility of disadvantageous cooperation is not surprising.' This paper identifies circumstances under which the possibility is likely. We study the case where countries that form the union (i.e. cooperate) coordinate trade policy in an attempt to alter the incentives faced by imperfectly competitive domestic industries. For simplicity we assume that these industries, each of which is treated as a single firm, produce for export only. We analyse three variations of a partial equilibrium trading model, which illustrate different reasons why limited cooperation may be worse than no cooperation for members of a union. We study complete information subgame- perfect equilibria. Section I considers the case where n firms, each identified with a different