Vocal learning in swamp sparrows, Melospiza georgiana, is subject to a host of sensory and motor limitations. One such limitation is that young swamp sparrows almost invariably crystallize their songs with a simple trilled syntax, irrespective of the syntax of vocal models from which they learn. A striking exception to this pattern was recently identified by Podos (1996, Animal Behaviour, 51, 1061-1070), who found that large-scale organizational changes in vocal syntax, including the production of an intermittent or 'broken' syntax, were produced when birds faced limits on vocal performance capacities during motor ontogeny. Our goal in the present study was to determine whether song models with broken syntax could serve as suitable training models for young swamp sparrows, and, if so, if broken syntax could be faithfully reproduced. We hand-reared 10 male swamp sparrows and exposed them to control, rapid and broken song models. Control song models were copied with a high degree of accuracy, as in previous studies. Rapid song models were copied with deficiencies that suggested performance limits on vocal production; such deficiencies included the production of songs with broken syntax and the production of songs in which notes were dropped out as songs progressed. Broken songs proved suitable as training models. Furthermore, copies of broken song models were crystallized either with normal or with broken syntax. These data identify an unexpected direction of permissiveness in the types of songs swamp sparrows will memorize and accurately reproduce, and also point to a possible proximate basis for syntactical changes in the evolution of sparrow songs. Copyright 1999 The Association for the Study of Animal Behaviour.