The 1961 Ellsberg paradox is typically seen as an empirical challenge to the subjective expected utility framework. Experiments based on Ellsberg's design have spawned a variety of new approaches, culminating in a new paradigm represented by, now classical, models of ambiguity aversion. We design and implement a decision-theoretic lab experiment that is extremely close to the original Ellsberg design and in which, empirically, subjects make choices very similar to those in the Ellsberg experiments. In our environment, however, these choices cannot be rationalized by any of the classical models of ambiguity aversion.