When a head-tracking system is coupled with a binaural rendering, the latter becomes dynamic. Above all, a headtracking system allows consistency between the orientation of the head and the audio rendering. This dynamism has many advantages, such as improved externalization, localization accuracy, or realism. However, an excessive latency between the head-tracking system and the audio rendering leads to a feeling of incoherence and/or slewing, which is detrimental to the listening experience. In this paper, the detection threshold of head-tracking latency was estimated using an adaptive procedure. This task was performed using three head rotation speeds and a total of six different sound scenes : a pink noise, a male speech with and without reverberation, a pair of congas with and without reverberation, and a complex scene (a realistic coffee shop ambiance, containing six distinct sources with reverberation).