We investigated whether measurement of pentosidine, in addition to the conventional risk assessment tool, the Fracture and Immobilization Score (FRISC), improves early identification of fracture cases. A total of 765 postmenopausal Japanese women with baseline measurement of urinary pentosidine were followed in a hospital-based cohort study. Endpoints were incidence of vertebral fracture, incidence of long bone fracture, and incidence of long bone and vertebral fracture. To assess the effect of pentosidine on fracture risk, we fitted multivariate Cox regression models adjusted for age, body weight, diabetes mellitus, lumbar BMD, prior fracture, and presence of back pain. To explore potential nonlinear relationships, we fitted a multivariate generalized additive model. To assess the discriminatory power of pentosidine, we performed receiver operating characteristic analysis. The hazard ratios for a 1 SD increase in pentosidine were 1.18 (95% CI 1.05-1.33, p < 0.01) for vertebral fracture and 1.20 (95% CI 1.07-1.33, p < 0.01) for long bone and vertebral fractures. The relationship was approximately linear, and there was no indication of the presence of a threshold. The C statistics were 0.732 (95% CI 0.686-0.778) for the model with both pentosidine and the 10-year risk and 0.702 (95% CI 0.654-0.750) for the 10-year risk alone. Eighty-three subjects (11%) in the whole cohort were in the highest quartile of pentosidine, although their 10-year risks were less than 15% and included 17 incident vertebral fracture cases. Urinary pentosidine improves risk classification using conventional risk assessment tools. Optimal clinical strategies of diagnosis and treatment remain uncertain and in need of additional investigation.