A phonological priming experiment reports inhibition for Russian prime-target pairs with onset overlap in native speakers. When preceded by the phonological prime /[image omitted]/, the target /kabak/ ("[image omitted]" -- "[image omitted]," "mare" -- PUB) takes longer to respond than the same target preceded by a phonologically unrelated word. English-speaking late learners of Russian also show inhibition, but only for high-frequency prime-target pairs. Conversely, they show facilitation for low-frequency pairs. In semantic priming (e.g. "carnation" -- "DAISY"), facilitation is observed for the same two lexical frequency ranges both in native speakers and learners of Russian, suggesting that the primes and targets in the low-frequency range are familiar to the nonnative participants. We interpret nonnative phonological facilitation for low-frequency words as evidence for sublexical processing of less familiar words that is accompanied by reduced lexical competition in nonnative lexical access. We posit that low lexical competition is due to unfaithful, or fuzzy phonolexical representations: nonnative speakers are unsure about the exact phonological form of low-frequency words. Such unfaithful representations are not strongly engaged in lexical competition and selection. High reliance on sublexical rather than lexical processing may be a general property of nonnative word recognition in case when the words are less familiar and have a low level of entrenchment. [Some of the material reported in the article is based upon work supported, in whole or in part, with funding from the United States Government.]