OBJECTIVE Current medication-adherence predictive tools are based on patient medication-taking beliefs, but studying past behavior may now be a more explanatory and accessible method. This study will evaluate if past medication-refill behavior for a statin regimen is more predictive of medication adherence than patient medication-taking health beliefs. METHODS This prospective longitudinal study was implemented in a national managed care plan in the United States. A group of 1433 statin patients were identified and followed for 6 months. Medication-taking health beliefs, collected from self-reported mail questionnaires, and past medication-refill behavior, using proportion of days covered (PDC), were collected prior to 6-month follow-up. Outcomes were measured using categorical PDC variable (of adherence, PDC ≥ 85%, versus nonadherence, PDC < 85%), with model fit estimated using receiver operator characteristic analysis. RESULTS The area under the receiver operator characteristic curve for past behavior (A(z) = 0.78) was significantly greater (P < 0.05) than for patient health beliefs (A(z) = 0.69), indicating that past prescription-refill behavior is a better predictor of medication adherence than prospective health beliefs. Among health beliefs, the factor most related to medication adherence was behavioral intent (odds ratio, 5.12; 95% confidence interval, 1.84 to 15.06). The factor most strongly related to behavioral intent was impact of regimen on daily routine (odds ratio, 3.3; 95% confidence interval, 1.41 to 7.74). CONCLUSIONS Electronic medical records and community health-information networks may make past prescription-refill rates more accessible and assist physicians with managing medication-regimen adherence. Health beliefs, however, may still play an important role in influencing medication-taking behaviors.