Data on expectations collected by business surveys offer the opportunity to test the rational expectation hypothesis. Tests directly based on qualitative data are not powerful because of the nature of the data. An alternative method is proposed in this paper. We apply the usual orthogonality conditions of rationality to the unobservable variables which are assumed to trigger the categorical responses of surveys when they cross certain thresholds. To implement the test, an original estimation method for latent variable models using indicators is required. The results show that the Hypothesis is not always rejected for the expectations of changes in demand drawn from surveys of French manufacturing industry.