A common form of phonological variation in American English is flapping. For example, the word ``pretty'' may be pronounced as [prIDi] or [prIti]. A phonological priming experiment was conducted where a lexical decision was made on a flap or canonical target. The target was preceded by itself (repetition condition), by its variant (form condition), or by a phonologically unrelated word (control). The results showed that priming for a canonical variant was comparable in the repetition and form condition. In contrast, priming for flaps was reduced in the form condition compared to the repetition condition. These findings suggest that the underlying voiceless stop is recovered when processing a flapped variant. The results are discussed in terms of surface versus underlying representations as the currency of lexical activation.