We study agents’ preferences for ambiguity resolution in dynamic environments. In an extension of the original Ellsberg experiment we first show the existence of preferences for late and gradual resolution of ambiguity. Most importantly we document a strong interdependence between agents’ attitude towards ambiguity and the preference for the timing and graduality of the resolution of ambiguity. We then compare our results to the predictions of popular recursive models of ambiguity aversion. Out of the considered ambiguity models only the smooth ambiguity model (Klibanoff Marinacci and Mukerji 2005) is flexible enough to accomodate the observed preferences.