Abstract Unlike most songbirds, individuals of several wood-warbler species have song repertoires that contain two song classes, each with distinct patterns of use and geographical variation. These patterns suggest that the modes of song development of the two categories could differ in important ways, such as in the nature of auditory and social experience required during song ontogeny. That possibility was tested in an experiment with a representative paruline species, the chestnut-sided warbler Dendroica pensylvanica. Of three experimental groups, one heard no tutor songs. A second group heard tutor tapes of both the unaccented-ending (UE) and accented-ending (AE) song categories, and a third group heard the tutor tapes but also saw and heard live adult males. The untutored birds developed no species-typical songs. Some of the tutored individuals learned songs from tapes alone, but learning and performance quality were greatly enhanced by the presence of a live tutor. Birds exposed only to tape tutoring imitated AE songs but failed to learn UE songs; live-tutored birds imitated songs from both categories. These results, interpreted in the context of field data, support the hypothesis that chestnut-sided warbler song development is somewhat compartmentalized, with AE song development more restricted by a neural ‘template’ and UE song more dependent on, and subject to influence by, social interaction.