ALMOST since the beginning of the systematic study of performance experimenters had noticed the tendency for subjects to combine ‘elements’ into ‘groups’ for perceptual and motor purposes. One of the classic studies in human learning is the series of experiments on morse-learning by Bryan and Harter1, who showed some sixty-five years ago that the process of learning was arrested periodically and that ‘plateaux’ appeared in the learning curve for receiving. This they interpreted by supposing that during the plateaux subjects consolidated what they had learned previously and reorganized elements into higher order groups: from dealing with inputs letter by letter one proceeds to words, phrases and whole sentences. A similar progression was postulated for learning to send, even though no plateaux fcwere found. Their interpretation has generally been accepted but so far as we know it has never been put to the test, We have gone a little way towards remedying this defect, and wish to communicate a first set of results.