The current study tested the assumption that bilinguals with dementia regress to using primarily the dominant language. Spanish-English bilinguals with probable Alzheimer's disease (AD; n = 29), and matched bilingual controls (n = 42) named Boston Naming Test pictures in their dominant and nondominant languages. Surprisingly, differences between patients and controls were larger using dominant-language than nondominant-language naming scores, and bilinguals with AD were either more likely than controls (in English-dominant bilinguals), or equally likely (in Spanish-dominant bilinguals), to name some pictures in the nondominant language that they could not produce in their dominant language. These findings suggest that dominant language testing may provide the best assessment of language deficits in bilingual AD, and argue against the common notion that the nondominant language is particularly susceptible to dementia. The greater vulnerability of the dominant language may reflect the increased probability of AD affecting richer semantic representations associated with dominant compared to nondominant language names. (Contains 4 tables and 3 figures.)