When testing for the existence of choice-induced changes in preferences, one is faced with the combined problem that the preferences are measured imperfectly and that the choice reflects the true preferences. The upshot is that the choice yields information about any measurement errors, implying that the choice may predict a change in measured preferences. Previous studies have neglected this effect, interpreting a change in measured preferences as a change in true preferences. This paper argues that the problems with previous studies can be mitigated by eliciting more information about the preferences of the participants prior to the choices. The paper reports results from a novel experiment, where the evidence does not support the existence of a choice-induced change in preferences.