Two studies examined language and understanding of scale models. First, children (N = 16; ages 2;4 to 3;5) received either the "standard" DeLoache model task or a "naming" version (in which children are asked to name the hiding location before retrieving a hidden object). Language ability positively correlated with performance on the model task, and children had significantly more errorless retrievals in the naming condition. Study 2 examined naming and transfer effects by giving children (N = 18; ages 2;5-3;1) both versions of the model task. Language ability correlated significantly only with performance in the naming condition. Both prior experience and naming condition significantly improved performance, with the effects being additive. Language and children's development of symbolic understanding are discussed. (Contains 1 table.)