The choice of grammatical framework in ICALL--the branch of CALL that applies artificial intelligence techniques--has important implications for both research and development. Matthews (1993) argued for one "that potentially meshes with SLA (second language acquisition)" (p. 5) and sketches three criteria that facilitate the crucial decision of selecting a grammatical framework for an ICALL system: computational effectiveness, linguistic perspicuity and acquisitional perspicuity. We will use Matthews' three adequacy criteria to review recent research in construction grammar and propose its application in ICALL, particularly in projects which involve the building of a student model. Such a student modeling project--Mocha--will be sketched briefly to provide a concrete context for the conceptualisation and implementation of construction grammar. This grammatical framework has the potential to help overcome some challenges in ICALL and to facilitate a more thorough analysis of learner language in context and thus improve our knowledge about language learning processes. (Contains 2 figures and 10 notes.)