OBJECTIVES: To explore changes in beliefs about medicines and self-reported medication non-adherence between 3 and 24 months after stroke and to investigate associations between beliefs about medicines and non-adherence at 24 months after stroke. DESIGN: Longitudinal questionnaire survey. SETTING: Patients treated for acute stroke in 25 Swedish hospitals. PARTICIPANTS: Only patients living at home were included. Of the 594 individuals who answered the 3 month questionnaire, 401 were included at 24 months; among the remainder, 34 (5.7%) had died, 149 (25,1%) did not respond or had incomplete information on adherence and 10 (1.7%) were not living at home. MEASURES: The primary outcome was self-reported medication adherence as measured with the Medication Adherence Report Scale (MARS). The Beliefs about Medicines Questionnaires (BMQ) was used to assess personal beliefs about medicines. Background and clinical data were included from the Swedish national stroke register. RESULTS: According to dichotomised MARS sum scores, more individuals were classified as non-adherent at 24 months after stroke (n=63, 15.7%) than at 3 months after stroke (n=45, 11.2%) (p=0.030). For BMQ, the only difference over time was an increase in the Necessity subscale (p=0.007). At 24 months, in comparison to adherent patients, non-adherent patients scored statistically significant higher on negative beliefs about medicines, such as Concern (OR 1.17, 95% CI: 1.09 to 1.25), Overuse (OR: 1.37, 95% CI: 1.21 to 1.54) and Harm (OR: 1.24, 95% CI: 1.11 to 1.39), and lower on positive beliefs about medicines, namely, Necessity (OR: 0.88, 95% CI: 0.80 to 0.96) and Benefit (OR: 0.85, 95% CI: 0.74 to 0.98). CONCLUSIONS: Stroke patients' beliefs about medicines were associated with adherence, and over time beliefs remained stable across all domains, except for an increased perception of medications as being necessary. Despite this, more patients became non-adherent over time. To counteract non-adherence, interventions targeted to improve intentional adherence as well as non-intentional adherence should be investigated and implemented.