A large research literature suggests that modelling in the absence of reinforcement for either the model's or the observer's behavior is a potent source of social learning. This literature is based entirely, however, on experiments using models that always display the critical behaviors. It is possible, therefore, that results obtained in these experiments would not generalize to natural settings in which modelling is intermittent. The effects of intermittent modelling were examined using three groups of 15 four‐ and five‐year‐old children. Male and female children from middle‐income families were individually exposed to an adult model who alternated descriptions of pictures of common objects. With one group, the model used no descriptive adjectives (color or number) in her descriptions after baseline; she used descriptive adjectives with 50% of the pictures with a second group, and 100% of the pictures with a third. Analyses of the data showed that the children substantially increased their use of descriptive adjectives in both modelling groups, but not in the no‐modelling group. The fact that there were no significant differences between the 50% and 100% modelling groups suggests that results obtained in studies using consistent modelling can be generalized to natural