The present study aims to isolate the locus of the frequency effect within the spoken word recognition architecture. By applying the additive factors logic (Sternberg, 1969) to an auditory lexical decision task where both word frequency and stimulus quality were factorially manipulated, the reaction time data can be analyzed to study processing stages along the time course of spoken word recognition, and determine if frequency has an early or late locus. A significant underadditive interaction of frequency and stimulus quality was obtained. Surprisingly, the typically robust frequency effect was not reliable for words of low stimulus quality. This finding suggests that word frequency influences a relatively late stage in the spoken word recognition process. Implications for extant models of spoken word recognition are discussed.