In this talk I will present work in progress on the cognitive mechanism for the representation of gradable adjectives. The use of gradable adjectives (such as tall, expensive, cold etc.) is agreed to be context-dependent (i.e., for different comparison classes the use of tall versus short, for example, will be different) and exhibits vagueness in the form of borderline cases (instances for which the speaker could use either of the antonym adjectives). Previous work on the representation and use of gradable adjectives use either a threshold model (e.g., Lassiter & Goodman, 2017; Qing & Franke, 2014) where the positive adjective is used to describe an instance if it is greater than a certain value, or a prototype model, where each new instance is categorized according to its distance from prototypical values of the positive and negative adjectives (e.g., Verheyen & Égré, 2018). In this study, we aim to test which of the two models best predict behavioural data and thus to better understand the categorization process in the representation of gradable adjectives.