Despite considerable debate about whether nonhuman primates learn to use tools via imitation, this type of learning by children has received surprisingly little attention. The findings of two studies that go some way toward filling this gap are reported here. Study 1 showed that when 2- and 3-year-old children (N= 68) were shown a correct solution to a tool-using task (which they could not solve spontaneously), all the children in both age groups managed at least a partial solution. When children were shown an incorrect solution followed by a correct solution, 2-year-olds again produced only a partial solution. By contrast, most 3-year-olds produced a full solution. Study 2 replicated this age change in a separate sample of children (N= 100) with a different tool-using task. Study 2 also showed that 3-year-olds benefit from observing an incorrect action when it can be contrasted with a correct action: they chose the more effective of the two actions. Taken together, the two studies indicate that by 3 years of age, children do not indiscriminately imitate actions on a tool, but selectively reproduce those actions that have a desired causal effect.