Many multisyllabic words contain shorter words that are not semantic units, like the CAP in HANDICAP and the DURA ("hard") in VERDURA ("vegetable"). The spaces between printed words identify word boundaries, but spurious identification of these embedded words is a potentially greater challenge for spoken language comprehension, a challenge that is handled by different mechanisms in different models of auditory word recognition. Subphonemic acoustic differences--subtle differences in pronunciation--often differentiate embedded words from genuine words. We examined semantic access to embedded words in two languages with different phonology by presenting carrier words followed by targets related to the embedded words and recording event-related potentials and lexical decision times in 34 Spanish/English bilinguals. No evidence of embedded word access was observed in brain activity or behavior, and this could not be attributed to subphonemic acoustic factors. The data place constraints on models of speech segmentation. (Contains 6 tables and 4 figures.)