A talker can sound quite different from one occasion to the next, and this has implications for the ease with which listeners can recognise that person from their voice alone. However, until relatively recently, natural within-talker variability was almost entirely overlooked in studies of voice identity perception. In this talk, I will present our work that has addressed how within-talker variability impacts voice identity perception, and how performance is affected by familiarity (e.g. with the talkers themselves, their accent, the language being spoken). Our approaches have included different perceptual paradigms (sorting, recognition, discrimination), manipulations of voice acoustics (e.g. F0 and formant spacing), and interrogation of the neural responses to natural stimulation (using multivariate analysis of fMRI data). I hope to convince the audience of the importance of including naturalistic variability in our studies of (speech and) voice processing, and will conclude by presenting some key questions for future research in this area.