Three studies investigatedchildren's ability to drawinferences fromtheproperties of one mental state to the properties of another. Inferences from knowledge/ignorance to the possible contents of pretends and beliefs are crucial to developing a representational theory of mental states. In Experiment 1, we replicated Lillard's (1993) finding that 4‐ and 6‐year‐olds fail to appreciate that a character who does not know about an entity cannot pretend to be that entity. We show that these children also fail a similar task in which the inference to be made is from not knowing to thinking (false belief). Lillard's inference tasks may be difficult because of their performance demands—specifically, children are not offered a plausible alternative content for the agent's pretence or belief state. In a second experiment, children were presented with know‐pretend and know‐think inference tasks which offered two options for the content of a character's mental state. One option was consistent with that character's knowledge state, while the other was not. Under these conditions, 4‐ and 6‐year‐oldchildren's performance improved significantly on both pretend and think. A third experiment investigated the role of the salience of the character's ignorance and the possible use of an association strategy in producing successful performance in Experiment 2. When the salience of the character's ignorance was reduced, children still succeeded on know‐pretend inferences but failed on know‐think inferences. These results suggest that children do not really grasp the theory of mental representation. The results better support the ToMM‐SP model of ‘theory of mind’ development. Accordingto this model, conceptpossession is prior to, and therefore does not depend upon, knowledge of theories, and task success depends upon the control of salience.