The elicited-response false belief task has traditionally been considered as reliably indicating that children acquire an understanding of false belief around 4 years of age. However, recent investigations using spontaneous-response tasks suggest that false belief understanding emerges much earlier. This leads to a developmental paradox: if young infants already understand false belief, then why do they fail the elicited-response false belief task? We postulate two systems to account for the development of false belief understanding: an association module, which provides infants with the capacity to register congruent associations between agents and objects, and an operating system, which allows them to transform these associations into incongruent associations through a process of inhibition, selection and representation. The interaction between the association module and the operating system enables infants to register increasingly complex associations on the basis of another agent's movements, visual perspective and propositional attitudes. This allows us account for the full range of findings on false belief understanding. (Contains 5 figures.)