Thick terms and concepts, such as honesty and cruelty, are at the heart of a variety of debates in linguistics, philosophy of language, and metaethics. Central to these debates is the question of how the descriptive and evaluative components of thick concepts are related and whether they can be separated from each other. So far, no empirical data on how thick terms are used in ordinary language has been collected to inform these debates. In this paper, we present the first empirical study, designed to investigate whether the evaluative component of thick concepts can be separated. Our study might be considered to support the view that separation is not possible. However, our study also reveals an effect of valence, indicating that people reason differently about positive and negative thick terms. While evaluations cannot be cancelled for negative thick terms, they can be for positive ones. Three follow-up studies were conducted to explain this effect. We conclude that the effect of valence is best accounted for by a difference in the social norms guiding evaluative language.