|Abstract:||The behavior of Russian [v] with respect to voicing assimilation and final devoicing presents a well known puzzle. Though [v] undergoes voicing assimilation and final devoicing, it fails to trigger voicing assimilation. The best known accounts for this problem posit that [v] is underlyingly /w/. These accounts presuppose (incorrectly) that all sonorants undergo voicing processes without triggering them. Further, they presuppose a kind of derivational opacity that cannot be recast in terms of any plausible model of derivational levels (as in Lexical Phonology) or output-output faithfulness. I offer a new account of Russian [v] that is entirely surface-oriented. This sound occupies a category intermediate between other obstruents and sonorants, something called a narrow approximant here. This fact, along with uncontroversial phonetic principles, explains why [v] behaves just as it does with respect to voicing processes. The account is supported by phonetic and typological observations, and turns out to be both empirically and explanatorily superior to previous accounts. An important consequence of the argument is that phonology must make reference to more phonetic distinctions than are made available by distinctive feature theory. I discuss how current approaches to contrast make this possible without overgenerating potential contrasts.