The term ‘uncanny valley’ was coined by Masahiro Mori in 1970 to describe the observation that near-human artefacts (such as a prosthetic hand) can trigger feelings of eeriness and repulsion in the viewer. A classic example is ‘Polar Express’, the 2004 animated feature film starring Tom Hanks, in which several of the characters (whose movements were derived using motion capture) are perceived by some observers as strange and uncanny. In science and engineering the uncanny valley has become of increasing relevance to character animation in computer games and to the appearance and behaviour of humanoid robots (such as Geminoid F), where the agents are sometimes perceived as being weird or creepy. Of course, the uncanny valley has been the subject of numerous scientific investigations. However, while some studies confirm its existence, others don’t find any evidence for it at all! Also, many of the theories about the effect are qualitative rather than quantitative. This paper shows, using a Bayesian model of categorical perception, how differential distortion caused by stimuli containing conflicting perceptual cues can give rise to tension in the region of a category boundary that serves to provide a potential quantitative explanation of the uncanny valley effect. The paper also shows how different observers could have different responses to the same stimuli, and concludes with some suggestions as to how the uncanny valley might be avoided in the design of future robots and autonomous agents by taking care to align the artefact's visual, vocal, cognitive and behavioural affordances.