Defenders of imprecise credences hold that ambiguous or unspecific evidence requires ambiguous or unspecific credences.  But there are some serious challenges to modeling rational agents using imprecise credences.  I offer a natural strategy for addressing some of these challenges. But adopting this strategy amounts to going precise, in a particular way: instead of attributing to the agent sets of credence functions, we attribute them uncertainty over sets of credence functions. I argue that the good reasons that have been offered for going imprecise are equally good, and sometimes better, reasons for adopting this kind of precise view.