We investigated associations of a broad biomarker panel with cognitive decline in atrial fibrillation (AF) patients to characterize possible mechanisms. We enrolled 1440 AF patients with available baseline biomarkers and cognitive testing by the Montreal Cognitive assessment (MoCA) score at inclusion and at ≥ 2 yearly follow-ups. We investigated the associations of biomarkers with cognitive decline in univariate logistic regression models, LASSO regression analysis and built a combined model. Mean age was 72 years, 75% male, 47% paroxysmal AF. Over 4 years, 93 patients (6.5%) had cognitive decline. These patients had more often permanent AF (32.3 vs 21.5%, p = 0.007) and more often a history stroke (23.7 vs 11.2%, p < 0.001), but similar baseline MoCA scores (24.9 vs 25.3 points, p = 0.22) and anticoagulation rates (93.5 vs 89.5%, p = 0.29). The three biomarkers with the highest univariate AUC for cognitive decline were GDF-15 (0.67 [0.62–0.72]), Cystatin C (0.67 [0.61–0.72]) and high-sensitivity Troponin T (hs-TnT) (0.65 [0.60–0.70]). In LASSO regression analysis, the best cross validation included GDF-15, GFAP, ESM-1, NfL and ALAT. The combined prediction model with the highest AUC of 0.73 (0.68–0.78) included IGFBP-7, GDF-15, Cystatin C, hsCRP, ALAT, GFAP, ESM-1 and FGF23. Over 4 years, 6.5% of AF patients had cognitive decline despite a high rate of anticoagulation. Inflammation, neuronal damage, and increased amyloid-beta might be important non-ischemic mechanisms of cognitive decline in AF patients.