Interpretations of longitudinal studies of cognitive aging are misleading unless effects of practice and selective drop-out are considered. A random effects model taking practice and drop-out into account analyzed data from four successive presentations of each of two intelligence tests, two vocabulary tests, and two verbal memory tests during a 17-year longitudinal study of 5,899 community residents whose ages ranged from 49 to 92 years. On intelligence tests, substantial practice effects counteracted true declines observed over 3 to 5 years of aging and remained significant even with intervals of 7 years between successive assessments. Adjustment for practice and drop-out revealed accelerating declines in fluid intelligence and cumulative learning, linear declines in verbal free recall, and no substantial change in vocabulary. Socioeconomic status and basal levels of general fluid ability did not affect rates of decline. After further adjustment for demographics, variability between individuals was seen to increase as the sample aged.