A long-standing hypothesis is that rapid automatized naming (RAN) measures access to phonological representations stored in long-term memory, but this has been difficult to test experimentally because phonological representations are mental constructs not easily operationalized. Here, we provide a method to test this theory using rime neighborhood density as an index of phonological representational strength. Thirty adults completed four picture-naming tasks orthogonalized for item composition (repeating vs. nonrepeating) and presentation format (discrete vs. serial). Each task was presented in two dichotomous conditions of rime neighborhood density (dense and sparse). There was no effect of rime neighborhood density on naming speed in the discrete nonrepeated (confrontation naming) task. However, rime neighborhood density significantly facilitated naming speed for serial repeated (i.e., RAN), discrete repeated, and serial nonrepeated tasks (ps<.03). The effect was weakest for confrontation naming (d=0.14) and strongest for both discrete and serial RAN tasks (ds=1.01), suggesting that repeating items, not serial presentation, makes RAN uniquely sensitive to manipulations of rime neighborhood density and, by proxy, phonological representations.