Evaluated the use of memory wallets (MWs) to self-prompt factual information during prompted conversations with familiar partners for 9 Ss (aged 67-93 yrs) with dementia. MWs contained 30 pictures and sentences about familiar persons, places, and events that each S had difficulty remembering. The amount of training by caregivers was the 1 component of the intervention package that varied among the 9 attempts to replicate condition effects observed on introduction of the wallets. All Ss learned to use the MWs to improve their conversations by making more accurate factual statements. Even when caregiver training was not provided to 3 Ss, condition effects were still demonstrated, with only 1 S requiring a brief orientation to his wallet. Long-term maintenance of condition effects was demonstrated for 3 Ss up to 30 mo post-intervention.