A battery of 17 neuropsychological tests (including the CERAD battery) and 17 psychometric ability tests were administered to a sample of 499 participants of the Seattle Longitudinal study who had been given the psychometric ability tests seven and 14 years earlier. The neuropsychological tests were projected into a 5-factor psychometric ability space by means of extension analysis. The concurrent regressions of the neuropsychology tests on the psychometric ability tests were then used to estimate neuropsychology test scores from the psychometric ability tests administered in 1984, 1991 and 1998. Neuropsychologists then rated the study participants as either normal, suspect or cognitively impaired in 1998. Changes in estimated test scores were computed over seven and fourteen years. Significant odds ratios between normal and cognitively impaired groups were found for all neuropsychological tests over the proximal period and for most tests over the 14-year period. Similar findings occurred for the odds ratios between the normal and suspect groups for the most proximal 7-year changes.