E. J. Robinson and P. Mitchell (1992) devised a message-desire discrepant task in which a speaker gives a message based on a false belief. A correct nonliteral interpretation of the message requires taking into account this false belief, which children were more likely to achieve in this task than in a classic prediction task. In 2 studies reported here, we repeated the comparison using more closely matched tasks. The message-desire discrepant task was no easier. In Study 3 we followed the Robinson and Mitchell procedure precisely but failed to replicate the contrast reported previously. Although the message-desire discrepant task reveals early understanding that messages are the product of mental representations, it offers no advantage over the classic test in revealing false-belief reasoning.