It is frequently assumed that the development of children's abilities in short-term memory reflects changes in a unitary short-term store. This approach makes only poor contact with recent research on adults, which suggests the idea of a more complex `working memory' system consisting of a limited-capacity central processor controlling a number of special-purpose stores. Two such stores are (i) the articulatory loop, a subsystem involved in subvocal rehearsal and associated with memory span, and (ii) the visuo-spatial scratch-pad, involved in imagery. This paper considers the applicability of the working memory framework to the study of children's memory. In adults, memory span for words is affected by their length, varying linearly with the rate at which they can be articulated, and thus presumably rehearsed. Studies of the developmental growth of memory span in children show that the same linear relation describes performance, with older children's better memory associated with faster rates of articulation. It appears from this that developmental change corresponds to an increase in the efficiency of subvocal rehearsal, with the decay characteristic of the articulatory loop remaining constant. However, although this simple developmental pattern is observed in memory for sequences of spoken words it is not present when the items are nameable pictures. Further investigation shows that older children use the articulatory loop to remember picture names: their performance is sensitive to phonemic similarity of the names and articulatory interference. However, younger children's performance is not affected by either of these factors but is sensitive to visual similarity. It is suggested that such children may be storing material in the visuo-spatial scratch-pad. An additional aspect of working memory is that separate mechanisms are thought to be involved in memory span and the `recency effect', the tendency for recent items in a list to be remembered well in unordered recall. A review of evidence obtained with children suggests that age differences in these two phenomena are independent. In general, therefore, it seems difficult to interpret the developmental changes reported here in terms of a unitary short-term store, and it is concluded that working memory provides a more promising approach.