This article addresses the thorny question of how to distinguish between complement and modifier PPs within the NP. Most of the accounts offered so far fail because they assume the distinction to be a rigid one: none of the semantic and syntactic criteria provided can be used as diagnostic tests for category membership, simply because the boundary between the two categories is not clear-cut. Instead, it is argued that the only viable way of accounting for the differences in formal behaviour between certain PPs within the NP is to recognize that this behaviour reflects a distinction at the conceptual level. Since, at the conceptual level, the distinction between ‘complement’ and ‘modifier’ is gradual (defined in terms of degree of activation), there is no basis for a strict linguistic dichotomy: therefore, the formal features of the PPs in question are to be regarded as approximations of the cognitive status of the concept in question, not as indications of all-or-none class membership. In addition to offering a more promising account of the complement–modifier distinction itself, the prototype-based approach proposed can also be regarded as providing plausible explanations for the formal differences commonly associated with (prototypical instances of) the two categories.