Overgeneralization occurs when a child uses the wrong word to name an object and is often observed in the early stages of word learning. We develop a method to elicit overgeneralizations in the laboratory by priming children to say the names of objects perceptually similar to known and unknown target objects. Experiment 1 examined 18 two-year-old children's labelling of familiar and unfamiliar objects, using a name that was previously produced. Experiment 2 compared the labelling of 30 two-year-olds and 39 four-year-olds when presented with completely novel objects. The findings suggest that the retrieved word is a blend of previous activation from the prior retrieval and activation engendered by the similarity of the test object to instances of the target category. We put forward a theoretical account of overgeneralization based on current models of adult language processing. The account suggests a common mechanism of activation and retrieval, which may explain not only momentary lapses in the correct selection of words, but other types of naming errors traditionally thought to reflect differences in children's underlying category representations or, perhaps, gaps in their knowledge of words.