Aims: Left atrial (LA) remodelling is a key determinant of atrial fibrillation (AF) ablation outcome. Optimal methods to assess this process are scarce. LA sphericity is a shape-based parameter shown to be independently associated to procedural success. In a multicentre study, we aimed to test the feasibility of assessing LA sphericity and evaluate its capability to predict procedural outcomes. Methods and results: This study included consecutive patients undergoing first AF ablation during 2013. A 3D model of the LA chamber, excluding pulmonary veins and LA appendage, was used to quantify LA volume (LAV) and LA sphericity (≥82.1% was considered spherical LA). In total, 243 patients were included across 9 centres (71% men, aged 56 ± 10 years, 44% with hypertension and 76% CHA2DS2-VASc ≤ 1). Most patients had paroxysmal AF (66%) and underwent radiofrequency ablation (60%). Mean LA diameter (LAD), LAV, and LA sphericity were 42 ± 6 mm, 100 ± 33 mL, and 82.6 ± 3.5%, respectively. Adjusted Cox models identified paroxysmal AF [hazard ratio (HR 0.54, P = 0.032)] and LA sphericity (HR 1.87, P = 0.035) as independent predictors for AF recurrence. A combined clinical-imaging score [Left Atrial Geometry and Outcome (LAGO)] including five items (AF phenotype, structural heart disease, CHA2DS2-VASc ≤ 1, LAD, and LA sphericity) classified patients at low (≤2 points) and high risk (≥3 points) of procedural failure (35% vs. 82% recurrence at 3-year follow-up, respectively; HR 3.10, P < 0.001). Conclusion: In this multicentre, real-life cohort, LA sphericity and AF phenotype were the strongest predictors of AF ablation outcome after adjustment for covariates. The LAGO score was easy to implement, identified high risk of procedural failure, and could help select optimal candidates. Clinical Trial Registration Information: NCT02373982 (http://clinicaltrials.gov/ct2/show/NCT02373982).