In a 5-year longitudinal study, elderly eminent academics were compared with elderly blue-collar workers on tests of intelligence and memory. Both blue-collar workers and academics deteriorated on non-verbal tasks. Their rate of decline did not differ. Contrary to expectations, blue-collar workers improved while academics deteriorated on a test of verbal reasoning; Moreover, a greater 'proportion' of the academics deteriorated on Similarities and the National Adult Reading Test (NART). In a cross-sectional analysis, the performance of the academics was compared with that of young Ph.D. students. With the exception of the NART, stem completion and figure reproduction, the mean scores of the academics were dramatically lower than that of the Ph.D. students. A few individuals performed above the mean level of Ph.D. students on each test but no individual performed above the mean on more than two of the four main tasks. Problems of regression to the mean, differential survival effects and sample size were addressed. Two conclusions were drawn: high ability is not associated with slower rates of decline and cognitive deterioration is universal on tests of non-verbal intelligence. (C) 1997 Elsevier Science Ltd.