Before children learn the specific meanings of numerals like six, do they know that they represent precise quantities? Previous studies have reported conflicting evidence and have found that children expect numerals to label precise quantities in some tasks but not in others (Condry & Spelke, 2008; Sarnecka & Gelman, 2004). In this article, we present evidence that some of children's apparent successes are best explained not by domain-specific semantic understanding of number but instead by language-general pragmatic abilities. In Experiment 1, we replicated the findings of the previous studies in a within-subject design. When 3-year-olds saw a set labeled with a number (e.g., "five") and an item was added, they preferred a new label ("six") over the old one, as though they believed that number words have precise meanings. However, when 1 of 2 sets was labeled (e.g., as "five") and children were asked to find the same quantity ("five") or a new quantity ("six"), they performed identically whether the original set was changed in quantity or merely rearranged. Thus, when 2 numerals were offered as alternative labels for 1 set, children behaved as though they had precise meanings, whereas when they were asked to determine which of 2 sets a single numeral referred to, they did not. In Experiment 2, children were tested using similar methods but with novel nouns and objects that were transformed, instead of sets. Children showed the identical pattern of results despite lacking meanings for these words, suggesting that their judgments for numerals may not have relied on semantic knowledge that numerals have precise meanings. We propose that children's behavior can be explained by the use of domain-general pragmatic inference and does not require positing domain-specific numerical knowledge. (Contains 6 figures and 3 footnotes.)