What semantics should we attribute to mass expressions like "wisdom" and "love", which are derived from gradable expressions? We first examine how these expressions are used, then how they are interpreted in their various uses. We then propose a model to account for these data, in which derived mass nouns denote instances of properties.