When readers or hearers interpret a definite associative NP, they must take into account that all definite NPs carry a presupposition of existential uniqueness. This implies that hearers or readers can access an entity which is presented as the only one of the type expressed by the N of the definite NP. In the case of associative definites which introduce a new entity, this entity must be easily accessible through, for instance, a definite NP 1 . Stereotypical part-whole relations are a case in point where such accessibility exists between the entities designated by NP 1 and NP 2 . The fact that a definite associative NP 2 must be supported by an accessibility relation explains why the use of a definite associative NP 2 comes across as strange, should a part-whole relation be transitive. Problems also arise when the hearer or reader has no previous knowledge about the links between the entities denoted by the definite NP 2 and the preceding NP 1 . In this case the context may offer several candidates for the role of antecedent and favor a coreferential or associative interpretation. The last part of the paper is devoted to a discussion of examples of this type of context and argues in favour of an approach to associative anaphora that takes into account diverse, and possibly contradictory, contextual clues in the utterance where NP 1 is employed.