Abstract Markers of bone turnover have been suggested to be useful in monitoring the long-term efficacy of antiresorptive therapy on bone mineral density (BMD). In this study, we developed a new model based on the combination of a marker level and its percent change at 6 months of therapy to predict long-term response in BMD. Serum bone alkaline phosphatase (BAP) was measured in 307 late postmenopausal women (mean age 64 years) with osteoporosis enrolled in a 2 year placebo-controlled trial of the bisphosphonate alendronate (10 mg/day). Under treatment, the maximal decrease was observed at 6 months (−44%) with no further change during the 2 year period. Both BAP levels at 6 months and percent BAP change at 6 months correlated with the percent change of spine BMD at 2 years ( r = −0.54 and −0.53, respectively, p < 0.001 for both). Logistic regression analysis showed that BAP levels and percent BAP change at 6 months are independent predictors of long-term positive BMD response, defined as ≥3% increase in spine BMD at 2 years. The most relevant clinical option that could lead to therapeutic adjustment is likely to be an accurate identification of nonresponders, and thus predictive models need to be highly specific. For a 90% specificity, the combination of both the percent change and BAP levels at 6 months resulted in a significantly ( p < 0.05) higher sensitivity (72%) than using percent BAP change (61%) or BAP level at 6 months (59%) alone. This combination model was also more effective than using the least-significant change (a decrease of BAP at 6 months of >44%) based on the within-patient variability in the placebo group. In the combination model, positive BMD responders vs. nonresponders could easily be distinguished by a line on a two-scale graph (BAP level at 6 month vs. percent BAP change at 6 months). In conclusion, the combination of BAP level and of its percent change after 6 months of treatment in a logistic model improved the prediction of the long-term BMD response to alendronate treatment compared with percent BAP change alone. This new model may be useful for quick and accurate identification of noncompliant patients (i.e., nonresponders) vs. responders to alendronate treatment, although prospective studies are required to determine accurately the rate of false positives and false negatives. Because this model is independent of the study design, it should be broadly applicable.