Despite the complexity and variation of physical signals, human perception of a speech sound uttered by different talkers or in diverse contexts is amazingly constant. Nonetheless, the neurocognitive mechanisms of this fundamental human perceptual ability are not well understood. Even less is known about the neural bases of phonetic constancy. We present an emerging picture of this important issue based on accumulating behavioral as well as neuroimaging evidence from lexical tone studies. Lexical tone offers a useful test of various existing theories proposed based on segmental studies, because of its variable and dynamic nature. We draw evidence from a series of studies on the perceptual normalization of lexical tones to shed light on prior theories. We also discuss findings from neuroimaging studies which advance our understanding of the temporal and spatial neural signature of phonetic constancy in lexical tone perception. A new model is proposed which emphasizes that listeners extract and integrate information from multiple sources in a fast and robust manner, to help them achieve phonetic constancy.