The practice of modelling behaviour using animats has met recent criticism. Since animats bear little structural resemblance to biological systems, it isn’t clear what animat models are models of. In this paper I identify and defend a central epistemic presupposition behind animat research; the claim that modellers can acquire new knowledge about biological systems without modelling their specific underlying mechanisms. I cast the critiques of animat research offered by Webb and Bechtel as forms of a general skeptical challenge to this presupposition and show how it could be met by an appeal to multiply-realisable behaviour types. I then outline a means of specifying such types dynamically.