This study empirically examines the relationships between union status, union involvement, and the performance of gain sharing programs. Using survey data from 217 establishments in the U.S.A. and Canada, the authors evaluate the predictions ofvarious competing theoretical perspectives: the agency/transaction cost approach, the monopoly model, the institutional voice model and a "two faces" model of labour organization. Ordered-probit and OLS regression results show that gain sharing programs with union involvement in program administration resulted in better perceived performance than average programs in the nonunion sector. However, gain sharing programs in the union sector without union involvement had worse outcomes than those in the nonunion sector. These two divergent situations resulted in union status itself having an insignificant relationship with program performance. These results are most compatible with the "two faces" model.