Intervocalic voicing neutralization has been generally accepted as a peculiar feature of Tohoku dialects. The present paper reports the results of perception experiments on this phenomenon. Natural and resynthesized stimuli spoken by Tohoku speakers were presented to both Tohoku and Tokyo listeners in a series of online experiments. A comparison between these two listener groups reveals that, for Tohoku listeners whose perception was biased by the voicing neutralization in their phonology, the boundary between voiced and voiceless tokens was more blurred compared to Tokyo listeners whose phonology had no such neutralization. These results suggest that neutralization can be bidirectional: i.e., voiced tokens become less voiced and voiceless tokens become less voiceless in contrast to the traditional view of neutralization which assumes a unidirectional process where one category remains intact and the other category merges with the former.