<p>Conceptual spaces are a frequently applied framework for analysing natural concepts. This paper explores the extent to which this analysis of natural concepts involves an interplay between descriptive and normative aspects—comparable to the way Bayesian cognitive science intertwines normative and descriptive elements in learning, reasoning, and argumentation. I argue that, although the two frameworks differ significantly in their general assumptions about core cognitive abilities (similarity-based versus reasoning-based), they share the use of normative assumptions as explanatorily valuable in understanding cognition. They also both face the question of realist commitments within a normative theory of belief (i.e., good credences reflect real-world statistics) and of concepts (i.e., good concepts as ‘carving <i>nature</i> at its joints’), respectively. I will argue that a complete account of natural concepts must remain open to such commitments.</p>