In speech comprehension, listeners recalibrate their interpretation of variable speech signals through exposure and disambiguating information. Recalibration is attested both segmentally and suprasegmentally, but little is known about what constrains it in lexical tone. This project investigated the effects of phonological categories and phonetic similarity on perceptual learning. We exposed Chinese listeners to pitch contours ambiguous between two tone categories (realised with a level or rising pitch contour) and lexically biased their perception to one interpretation. Crucially, the rising pitch contour could be from two different phonological tone categories. Perceptual learning was observed not only in the rising contours used for exposure but also across phonetically similar but phonologically different rising contours, suggesting that perceptual learning in tone is not constrained by phonological tone categories but is facilitated by the phonetic similarity of pitch contours.