Abstract One of the major problems in the estimation of disequilibrium models is to obtain a regime repartition which is coherent with a priori findings. The paper proposes a Bayesian analysis of this type of models in order to be able to introduce prior information on the regime repartition and on the structural parameters. Our main concern is to study the internal logic of the prior density. In particular coherence between two sources of information (business survey data and economic theory) is obtained by a procedure of selection on the conditioning variables of the model. The practical working of the method is shown on an economic example.