Relatively few systems for machine hearing exploit top-down information in source localisation, despite there being clear evidence for top-down (e.g., attentional) effects in biological spatial hearing. This paper addresses this issue by proposing a framework for binaural sound localisation that exploits top- down knowledge about the source spectral characteristics in the acoustic scene. Information from source models is used to improve the localisation process by selectively weighting binaural cues. The system therefore combines top-down and bottom- up information flow within a single computational framework. Our experiments show that by exploiting source models in this way, sound localisation performance can be improved substantially under challenging conditions in which multiple sources and room reverberation are present.