Although biochemical markers of skeletal turnover cannot replace bone density scanning for the diagnosis of osteoporosis, it is thought that they may help add to prediction of fracture risk and help determine adequacy of osteoporosis therapy. Nevertheless, whether biochemical markers in the serum or urine can predict individual rates of bone loss in the spine or hip region is unknown. We studied a heterogeneous group of women (n = 81) who were premenopausal, untreated postmenopausal, and estrogen-treated postmenopausal with baseline determination of body mass index (BMI), calcium intake, biochemical measurements, and serial bone densitometry over 3 years. Serum assays included bone Gla protein (BGP), total and bone-specific alkaline phosphatase (AP, BSAP), carboxyterminal propeptide of type I procollagen (PICP), carboxyterminal telopeptide of type I collagen (ICTP) and tartrate-resistant acid phosphatase (TRAP). Urine assays included hydroxyproline (OHP), calcium, total pyridinoline, and total deoxypyridinoline. Individual biochemical markers and calcium intake were modestly correlated with bone density changes but were inconsistent regarding the spine versus the hip. All of the formation variables were significantly correlated to spine density change (r = -0.24 to -0.49) whereas the only resorption variable that correlated was urine OHp/Cr (r = -0.31). The only formation variable that correlated with hip density change was serum PICP whereas all of the resorption variables except serum TRAP were correlated (r = -0.23 to -0.35). "High turnover" individuals were defined at those with levels of biochemical variables at least 1 SD above the mean young normal for each variable. Higher bone loss rates were seen in this group for several of the turnover markers compared with bone loss rates in all other individuals. However, the sensitivity of this "high turnover" status for identifying high bone losers did not exceed 60% for any of the variables. In untreated postmenopausal women, a model using urine OHp, serum ICTP, serum BSAP, and calcium intake was able to predict 42% of the variance of change in BMD of the lumbar spine. A model using BMI, serum ICTP, and serum BGP could predict 32% of the variance of change in BMD of the femoral neck. No combination of markers could predict variance in bone density change at either site in estrogenized women (premenopausal and estrogen-treated postmenopausal). We conclude that measuring individual serum and urine markers of bone turnover cannot accurately predict bone loss rates in the spine and hip; however, combinations of demographic and biochemical variables could predict some of the variance in untreated postmenopausal women. Biochemical markers cannot replace serial bone densitometry for accurate determination of change in bone mass at the most clinically relevant sites.