The present study examined how differences in onset (cohort) and offset (rhyme) neighborhood density influence the types of spoken word recognition errors made by listeners. Simulations of the TRACE model were used to derive preliminary predictions. Younger (N=15) and older (N=15) adults identified spoken words presented in moderate noise. Participants exhibited the standard inhibitory effect of phonological neighborhood density: slower recognition of spoken words from denser neighborhoods, with a larger effect for older adults. Most errors were phonological neighbors with few unrelated errors. However, the manipulation of cohort and rhyme density produced an unexpected reversal: the relative proportion of cohort vs. rhyme errors was biased toward cohorts when cohort density was low or when rhyme density was high, and toward rhymes when cohort density was high or rhyme density was low. These results are not consistent with the TRACE simulations and suggest a more complex pattern of lexical competition