Studies of wild capuchins suggest an important role for social learning, but experiments with captive subjects have generally not supported this. Here we report social learning in two quite different populations of capuchin monkeys (Cebus apella). In experiment 1, human-raised monkeys observed a familiar human model open a foraging box using a tool in one of two alternative ways: levering versus poking. In experiment 2, mother-raised monkeys viewed similar techniques demonstrated by monkey models. A control group in each population saw no model. In both experiments, independent coders detected which technique experimental subjects had seen, thus confirming social learning. Further analyses examined fidelity of copying at three levels of resolution. The human-raised monkeys exhibited fidelity at the highest level, the specific tool use technique witnessed. The lever technique was seen only in monkeys exposed to a levering model, by contrast with controls and those witnessing poke. Mother-reared monkeys instead typically ignored the tool and exhibited fidelity at a lower level, tending only to re-create whichever result the model had achieved by either levering or poking. Nevertheless this level of social learning was associated with significantly greater levels of success in monkeys witnessing a model than in controls, an effect absent in the human-reared population. Results in both populations are consistent with a process of canalization of the repertoire in the direction of the approach witnessed, producing a narrower, socially shaped behavioural profile than among controls who saw no model.