To determine the ability of bone density to predict fractures in very elderly women, we examined the association of bone density with non-spine, hip, wrist and humerus fractures in 8699 Caucasian women aged 65-79 years and 1005 women aged 80 years and older who were participants in the prospective Study of Osteoporotic Fractures. Follow-up averaged 4.9 (+/- 1.0) years after baseline measurement of appendicular bone density (single-photon absorptiometry; Osteon-Osteo Analyzer) and 2.9 (+/- 0.6) years after measurement of hip and anteroposterior lumbar spine bone density (dual-energy X-ray absorptiometry; Hologic QDR 1000) at a later examination. In general, measurements of bone density in the oldest women retained their predictive value for fractures. Among women aged 80 and over a 1 standard deviation decrease in bone density of the distal radius was associated with an increased risk of non-spine (relative risk: 1.6; 95% confidence interval: 1.3, 1.8), wrist (1.7; 1.2, 2.4) and humerus fracture (2.9; 1.8, 4.7), while a 1 standard deviation decrease in femoral neck bone density was associated with an increased risk of non-spine (1.9; 1.6, 2.4), humerus (2.4; 1.2, 4.7) and hip fracture (2.1; 1.4; 3.2). However, only trochanteric but not femoral neck fractures were associated with low bone density in these oldest women. The excess risk of fracture in women with below-median bone density was greater in those aged 80 years and over compared with the younger women: 38.4 v 20.4 per 1000 woman-years for non-spine fracture and 12.2 v 3.2 per 1000 woman-years for hip fracture.(ABSTRACT TRUNCATED AT 250 WORDS)