Background. We report on the change in cognitive function in a populat
ion sample of elderly people who have been examined on two occasions m
ore than 3 years apart. Methods. A sample of 1135 persons aged 70-102
years was interviewed at base-line then reinterviewed 36 years later w
ith the Canberra Interview for the Elderly, which included tests of ep
isodic memory and cognitive speed as well as the Mini-Mental State Exa
mination and the National Adult Reading Test (NART). Results. Mortalit
y and loss to follow-up reduced the sample to 736, of whom 614 complet
ed at least one test of cognitive performance on both occasions. Cogni
tive performance decreased with age, except on the NART. Decline over
the follow-up period increased as a function of age in all cognitive m
easures, except the NART. Change in cognitive scores was close to norm
al distribution. Incident dementia was associated strongly with age an
d current level of cognitive performance, but not with rate of decline
. Cognitive decline and the risk of incident dementia did not differ b
y gender. Conclusions. A score indicating possible impairment in the v
ery elderly carries a worse prognosis than for the younger elderly. De
cline is almost universal in at least one cognitive area among those o
ver the age of 85.