A number of contradictory theoretical hypotheses have been advanced about the relationship between unionisation and job satisfaction. In this article, new evidence of the effects of unionisation on job satisfaction is presented using French linked employer–employee data from the 2011 REPONSE Survey. A bivariate probit model is estimated to deal with the reverse causation issue that many previous studies have failed to account for. The results indicate that union members are less satisfied with their jobs than non-members, even after taking into account certain individual characteristics, job attributes and workplace characteristics. However, after controlling for endogeneity of union membership, we find that the difference in job satisfaction between unionised and non-unionised workers disappears because of a selection effect in workplaces covered only by a national/sectoral collective agreement and attributable to a causal effect arising from union's voice function in workplaces covered by a local agreement.