Some of the most compelling applications of spoken language technology in education involve children, but computer recognition of children’s speech is particularly difficult. This paper reviews current approaches to children’s speech recognition. It concludes that a fundamental challenge is to raise the performance of ‘matched’ children’s speech recognition systems to the same levels that are achieved by state-of-the-art systems for adults, or, alternatively, to explain why this is not possible, for example by studying human recognition of children’s speech. It is suggested that to achieve this it will be necessary to take account of the stages of acquisition and development of children’s speech. These processes are well documented in the speech therapy literature, but their computational utility is still to be demonstrated.