There has been extensive research on the use of both stand-alone and embedded measures of effort in neuropsychological testing; however, relatively few studies have reported on their utility in the context of dementia. Previous studies that have examined the specificity of traditionally used cut-scores on embedded measures of effort with dementia samples have largely found high rates of false positive errors. The present study examined the specificity of several Digit Span derived embedded measures of effort in a large clinical sample of patients with probable Alzheimer's disease stratified by level of dementia severity. Of the measures that were examined, only the Vocabulary - Digit Span score demonstrated promising specificity through the moderate level of dementia severity. All of the remaining indices, including Reliable Digit Span, Digit Span Age-Corrected Scaled Score, and Longest Digits Forward (1 & 2 Trials), yielded unacceptable rates of false positive errors as dementia severity increased. The implications for these findings are discussed, including the limitations of importing methods of assessing effort from one sample to another.