Reliable prediction of very late recurrence of atrial fibrillation (VLRAF) occuring >12 months after catheter ablation (CA) in apparently "cured" patients could optimize long-term follow-up and modify decision-making regarding the discontinuation of oral anticoagulant therapy. In a single-centre cohort of consecutive patients post radiofrequency AFCA, we retrospectively derived a novel score for VLRAF prediction. Of 133 consecutive post AFCA patients (mean age 56.9 ± 11.8 years, 63.9% male, 69.2% with paroxysmal AF) who were arrhythmia-free at 12 months (excluding 3-month "blanking period"), 20 patients expirienced a VLRAF during a 29.1 ± 10.1-month follow-up, with a 3-year cumulative VLRAF rate of 31.1%. The MB-LATER score (Male, Bundle brunch block, Left atrium ≥47 mm, Type of AF [paroxysmal, persistent or long-standing persistent], and ER-AF = early recurrent AF), had better predictive ability for VLRAF (AUC 0.782) than the APPLE, ALARMc, BASE-AF2, CHADS2, CHA2DS2VASc or HATCH score (AUC 0.716, 0.671, 0.648, 0.552, 0.519 and 0.583, respectively), resulted in an improved net reclassification index (NRI) of 48.6-95.1% and better identified patients with subsequent VLRAF using decision-curve analysis (DCA). The MB-LATER score provides a readily available VLRAF risk assessment, and performs better than other scores. Validation of the MB-LATER score in other cohorts is underway.