As an emerging form of intermodal translation, audio description (AD) raises many new questions for Translation Studies and related disciplines. This paper will investigate the question of how the coherence of a multimodal source text such as a film can be re-created in audio description. Coherence in film characteristically emerges from links within and across different modes of expression (e.g., links between visual images, image-sound links and image-dialogue links). Audio describing a film is therefore not simply a matter of substituting visual images with verbal descriptions. It involves ‘translating’ some of these links into other appropriate types of links. Against this backdrop, this paper aims to examine the means available for the re-creation of coherence in an audio described version of a film, and the problems arising. To this end, the paper will take a fresh look at coherence, outlining a model of coherence which embraces verbal and multimodal texts and which highlights the important role of both source text author (viz., audio describer as translator) and target text recipients in creating coherence. This model will then be applied to a case study focussing on the re-creation of various types of intramodal and intermodal relations in AD.