Most current accounts of color word acquisition propose that the delay between children’s first production of color words and adult-like understanding is due to problems abstracting color as a domain of meaning. Here we present evidence against this hypothesis, and show that, from the time children produce color words in a labeling task they use them to represent color. In Experiment 1, an analysis of early color word production errors finds that, before acquiring adult-like understanding, children make systematic hypotheses about color word meanings, which are best characterized as overextensions of adult meanings. Experiment 2 analyzes comprehension errors and finds that these overextensions result from overly broad categories, rather than a communicative strategy. These results indicate that the delay between production and adult-like understanding of color words is largely attributable to the problem of determining language-specific color boundaries.