Adherence to anti-osteoporotic regimens gradually decreases over time. We hypothesized that the determinants of non-compliance or non-persistence at different times vary and identified these differences. We used an outpatient database to retrieve information on anti-osteoporotic medications prescribed by a medical centre in southern Taiwan during 2001-2007. Compliance was defined as a medication possession ratio (MPR) ≥80 %. Persistence was determined as continuous use, allowing for a refill gap of 30 days. A multivariate Cox regression model evaluated potential predictors of non-adherence. A total of 3589 patients were included. In the multivariate analyses, non-compliance for both year 1 and year 2 was more likely in patients with non-vertebral non-hip fractures, respiratory disorders, prescription of the first anti-osteoporotic regimen by an orthopedist; and less likely in patients with follow-up bone densitometry and switched regimens. Risks for non-persistence at year 1 and year 2 were generally similar to those for non-compliance; insurance coverage and malignancy were associated with a lower risk of non-persistence at year 1 and year 2, respectively. In the subgroup with an MPR ≥80 % at year 1, an index prescription by an orthopedist was the only independent predictor of non-compliance and non-persistence at year 2. In conclusion, the positive or negative determinants of non-adherence were different at year 1 and year 2, which indicated that clinicians might deliver effective interventions to improve adherence via different precautions annually. This study also provided evidence that physician specialty had a significant effect on adherence to osteoporosis care.