The experimental study of decision-making under uncertainty typically builds on Ellsberg's (1961) setting. Yet, as the total number of balls is known in standard Ellsberg's urns, an implicit constraint is put on the specification of the probability models to consider. In practice, this restricts the ability of Ellsberg's urns to characterize situations going beyond those of model ambiguity. In this note, I present a simple and easy-to-implement device that creates the initial conditions of uncertainty, which constitute a critical prerequisite for the study of model misspecification.