Learners frequently experience phonologically inconsistent input, such as exposure to multiple accents. Yet, little is known about the consequences of phonological inconsistency for language learning. The current study examines vocabulary acquisition with different degrees of phonological inconsistency, ranging from no inconsistency (e.g., both talkers call a picture /vig/) to mild but detectable inconsistency (e.g., one talker calls a picture a /vig/, and the other calls it a /vIg/), up to extreme inconsistency (e.g., the same picture is both a /vig/ and a /dIdʒ/). Previous studies suggest that learners readily extract consistent phonological patterns, given variable input. However, in Experiment 1, adults acquired phonologically inconsistent vocabularies more slowly than phonologically consistent ones. Experiment 2 examined whether word-form inconsistency alone, without phonological competition, was a source of learning difficulty. Even without phonological competition, listeners learned faster in 1 accent than in 2 accents, but they also learned faster in 2 accents (/vig/ = /vIg/) than with completely different labels (/vig/ = /dIdʒ/). Overall, results suggest that learners exposed to multiple accents may experience difficulty learning when 2 forms mismatch by more than 1 phonological feature, plus increased phonological competition due to a greater number of word forms. Implications for learning from variable input are discussed.