In this article we present two sets of experiments designed to investigate the acquisition of scalar implicatures. Scalar implicatures arise in examples like Some professors are famous where the speaker’s use of some typically indicates that s/he had reasons not to use a more informative term, e.g. all. Some professors are famous therefore gives rise to the implicature that not all professors are famous. Recent studies on the development of pragmatics suggest that preschool children are often insensitive to such implicatures when they interpret scalar terms (Noveck 2001 for terms like might and some; Chierchia, Crain, Guasti, Gualmini and Meroni 2001 for or). This conclusion raises two important questions: a) are all scalar terms treated in the same way by young children?, and b) does the child’s difficulty reflect a genuine inability to derive scalar implicatures or is it due to demands imposed by the experimental task on an otherwise pragmatically savvy child? Experiment 1 addresses the first question by testing a group of 30 5-year-olds and 30 adults (all native speakers of Greek) on three different scales, meriki/oli (some/all), dio/tris (two/three) and arxizo/teliono (start/finish). In each case, subjects were presented with contexts which satisfy the truth conditions of the stronger (i.e. more informative) terms on each scale (i.e. all, three and finish) but were described using the weaker terms of the scales (i.e. some, two, start). We found that while adults overwhelmingly rejected these infelicitous descriptions, children almost never did so. Children also differed from adults in that their rejection rate on the numerical scale was reliably higher than on the two other scales. In order to address question (b), we trained a group of 30 5-year-olds to detect infelicitous statements. We then presented them with modified versions of the stories of Experiment 1, which now more readily invited scalar inferences. These manipulations gave rise to significantly higher rejection rates than those observed in Experiment 1. Overall, these findings indicate that children do not treat all scalar terms alike and, more importantly, that children’s ability to derive scalar implicatures is affected by their awareness of the goal of the task. Developmental and methodological implications as well as theoretical implications for the semantics of numeral terms are discussed.