Children's understanding of density is riddled with misconceptions--or so it seems. Yet even preschoolers at times appear to understand density. This article seeks to reconcile these conflicting outcomes by investigating the nature of constraints available in different experimental protocols. Protocols that report misconceptions about density used stimulus arrangements that make differences in mass and volume more salient than differences in density. In contrast, protocols that report successful performance used stimulus arrangements that might have increased the salience of density. To test this hypothesis, the present experiments manipulate the salience of object density. Children between 2 and 9 years of age and adults responded whether an object would sink or float when placed in water. Results indicated that children's performance on exactly the same objects differed as a function of the saliency of the dimension of density, relative to the dimensions of mass and volume. These results support the idea that constraints--rather than stable knowledge--drive performance, with implications for teaching children about nonobvious concepts such as density. (Contains 8 figures and 4 footnotes.)