Objective: To perform a long-term evaluation of the localisation capabilities in the horizontal plane of single-sided deaf patients fitted with a BAHA device. Design: Single-centre retrospective study. Participants: Twenty-one adults with single-sided deafness (SSD) with normal hearing in the contralateral ear (pure tone average <20 dB, SDS > 90%) rehabilitated with a Cochlear BAHA device from 2003 to 2012 on the deaf side over a median follow-up of 8 years. Outcome measures: The task used in this paper is a sound localisation identification task with a set-up of seven loudspeakers on a semi-circular array at 30-degree intervals performed at three periods: before BAHA, initially and at last follow-up. Our main criterion of judgement was the root-mean-square (RMS) localisation error. In addition, the Bern Benefit in Single-Sided Deafness Questionnaire (BBSS) was administered. Results: The mean RMS localisation error was initially estimated at 64° without any rehabilitation (for a chance level RMS estimated at 81°). Initially, with the BAHA device, the RMS localisation error dropped to 51°. At the last follow-up evaluation, a significant decrease at 23° was noted. Concerning the Bern Questionnaire, 19% of the patients (n = 4) did not report any change (score of 0), 33% (n = 7) are satisfied (score of +1 or +2) and 48% (n = 10) are very satisfied with the BAHA device (score better than +3). Conclusion: Improvement of sound localisation in the horizontal plane for some SSD patients is likely related to altered processing of monaural spectral cues. The time needed to learn to use the azimuth-dependent spectral cues takes time. Long-term follow-up should be considered for studies investigating sound localisation performance.