This article proposes a formal model of the human language faculty that accommodates the possibility of 'attrition' (modification or loss) of morphosyntactic properties in a first language. Modelling L1 performance factors are more immediately perceptible both to attriters themselves and to researchers, whereas the properties of speakers' underlying grammars are typically below the level of consciousness. Moreover, the potential for attrition of morphosyntactic properties beyond childhood appears to be highly restricted and is considerably rarer, as is frequently reported in the literature (see, e.g., reviews in Domínguez 2013; Tsimpli 2017; Schmid and Kӧpke 2017a,b).While stable L1 grammars appear unlikely to be significantly restructured beyond childhood, grammatical attrition-in the sense defined above-is indeed evidenced in the literature, for example in the interpretation of overt pronouns (e.g. Sorace 2000; Tsimpli, Sorace, Heycock and Filiaci 2004; Tsimpli 2007, Domínguez 2013) and properties of pronominal binding (e.g. Gürel 2004Gürel , 2007. While not as widely attested as attrition in L1 performance or processing, we contend that from a scientific perspective, the possibility of grammatical attrition is crucial to our understanding of bilingual acquisition and of the organisation of the language faculty, broadly speaking. However, to our knowledge there is currently no model of grammatical attrition compatible with formal generative models of the language faculty. 3More problematically still, any formal approach to modelling grammatical attrition faces an apparent paradox: if the language faculty architecture comprises mechanisms capable of affecting a mature L1 grammar, then why is the phenomenon of grammatical attrition itself so heavily constrained and so apparently rarely attested? Any model seeking to accommodate attested instances of grammatical attrition will then not only need to account for what may be modified within the L1 grammar (and under what conditions), but will also need to resolve this paradox in modelling attrition.The principal aim of this article is to articulate such a model. Section 2 first establishes a set of requirements which we argue any model of L1 grammatical attrition must meet. We then articulate our own model of L1 grammatical attrition incorporating three key aspects: first, a distinction between input and intake (as input which has been processed and assigned a representation, e.g. Carroll 2001); second, an inference component that modifies an existing grammar under appropriate intake conditions, including parsing and extralinguistic factors such as memory and pattern recognition (e.g. Lidz and Gagliardi 2015); finally, a 'featurebased' generative model of the computational component of the grammar prominent in the theoretical literature (e.g. Chomsky 2000(e.g. Chomsky , 2001. The basis of our model is provided by Lidz and Gagliardi's (2015) L1 acquisition model, as a framework for understanding how input may engender changes to the state of an L1 grammar provided i...