Two experiments investigating the capacity to sustain mental set in dementia were conducted. Experiment 1 analyzed performance of a non-demented control group (NC), participants with Alzheimer's disease (AD) and participants with ischemic vascular dementia (IVD) on the Boston Revision of the Wechsler Memory Scale Mental Control subtest (MC). On simple tasks there were no between-group differences after controlling for time to completion. On complex tasks, NC participants outperformed both dementia groups and AD participants obtained higher accuracy indices than IVD participants. The IVD group produced a disproportionate number of commission errors regardless of task complexity. The AD group tended to produce more omission errors on more difficult measures of mental set. Individual task performance was divided into three sections-first, middle, and last. IVD participants made fewer and fewer correct responses over all three sections, whereas performance of AD participants leveled off by the middle section with no further decline. Experiment 2 compared letter fluency performance among NC, AD and IVD groups, and participants with dementia secondary to idiopathic Parkinson's disease (PD). For all letter cues, IVD and PD participants generated fewer responses than NC and AD participants. However, IVD and PD participants generated a larger proportion of words than AD and NC participants within the first 15 s. As the task progressed, the output of IVD and PD participants dropped precipitously. These findings indicate that failure to maintain mental set is not a diffuse or general cognitive disability. Rather, failure to maintain mental set in dementia may be best understood within the context of predictable and specific within-task time epochs.