Objective: To learn how pAD (probable Alzheimer's disease), PD+ ('Parkinson's Plus' syndrome), and control subjects remember internally generated material under different conditions. Background: 'Self-discovered,' or internally generated knowledge, prized by educators and therapists, can bring about considerable behavioral change. Both parietal-temporal-limbic (pAD) and frontal-subcortical dementia (e.g. PD+) cause dysmemory, but may cause different internal-external memory bias. pAD subjects, confusing internal and external information (confabulation) and reporting internal information during memory testing (intrusions), may be biased to remember internal material. PD+ subjects, impaired at generative tests, may be externally biased. Methods: Ten pAD, 5 PD+, and 10 control subjects generated words in a category without instruction to remember (INR), and took a list- learning test of incidental memory for internally and externally generated words. To test how INR influences memory, subjects then generated and attempted to recall four more words. Results: All three subject groups remembered more internally generated than externally provided words without INR. Recall versus recognition of internally generated words differed by group, with PD+ subjects showing greatest improvement with recognition. The pAD subjects performed worse with INR than without INR, had the most intrusion errors, and, rather than demonstrating a release from proactive inhibition, recalled fewer words outside the category. Groups differed in overall recall/recognition improvement (p = 0.015). Conclusions: Aged subjects preferentially retained internally generated material. However, among demented subjects, memory for internally generated words was influenced by the testing method used. PD+ subjects have poor internal recall, but excellent internal recognition. In pAD, memory for internally generated words may exceed external memory, but only when subjects are not explicitly trying to remember.