Lexical tone languages make up the majority of all known languages of the world, but the role of tone in lexical processing remains unclear. In the present study, four form priming experiments examined the role of Mandarin tones in constraining lexical activation and the time course of the activation. When a prime and a target were related directly in form (e.g., lou3 'hug'--lou2 'hall'), competitors that differed from the prime in tone failed to be activated, indicating the use of tonal information to distinguish between segmentally identical words. When a prime and a target were not form-related but were related through a third word that was not actually presented (e.g., lou3 'hug'--jian4zhu0 'building', where lou3 is form-related to lou2 'hall', which was semantically related to jian4zhu0), a mismatch in tone prevented activation of minimal tone pairs at 250ms interstimulus interval (ISI) but did not prevent activation at 50 ms ISI. These results indicate that tonal information is used on-line to reduce the number of activated candidates, but does not prevent the minimal tone pairs from being activated in the early phase of lexical activation.