The acoustic duration of the English past tense (ED) morpheme was measured for matched high‐ and low‐frequency verbs (e.g., KNEADED/NEEDED). The ED on low‐frequency verbs was of longer acoustic duration than the ED on matched high‐frequency words. The rhyming portion of the matched verbs also showed a lengthening effect for the low‐frequency words; in contrast to previous reports that word frequency does not affect word acoustic duration [Geffen and Luszcz, Mem. Cogn. 11, 13–15; Wright Mem. Cogn. 7, 411–419]. However, this effect was statistically independent of the ED length effect, and the final phonetic segment of a low‐frequency monomorphemic verb stem was not longer in acoustic duration than a homophonous segment on a matched high‐frequency verb (e.g., the /d/ in KNEAD/NEED). Further, the ED morpheme is of longer acoustic duration than a homophonous segment in a nonverb homophone (e.g., RAPPED/RAPT), as earlier reported for the morpheme /s/ [Walsh and Parker, J. Phon. 11, 201–206]. This set of evidence corroborates a frequency‐dependent dual‐access processing theory of linguistic morphology: that high‐frequency complex words are processed holistically and low‐frequency complex words are processed componentially [cf. Bybee, Morphology (1985)].