The Cox maze procedure was introduced in 1987 for the treatment of atrial fibrillation. This study evaluated the predictors of late atrial fibrillation recurrence in 276 consecutive patients who underwent this procedure at our institution. From 1987 through June 2003, 276 patients (79 female and 197 male patients; mean age, 55 ± 11 years) underwent the Cox maze procedure. Thirty-three patients had Cox maze procedure I, 16 patients had Cox maze procedure II, and 197 patients had Cox maze procedure III. The last 30 patients underwent a modified procedure (Cox maze procedure IV) with bipolar radiofrequency ablation. There were 113 (41%) patients who had a concomitant operation, most commonly either a mitral valve procedure (19%) or coronary artery bypass grafting (20%). Data were analyzed by means of univariate analysis, with preoperative and perioperative variables used as covariates. Patient follow-up was conducted by means of questionnaire, physician examination, and electrocardiographic documentation. All patients had a minimum of 6 months of follow-up. Patient follow-up was achieved in 92.8% of cases, with a mean follow-up time of 5.8 ± 3.6 years. Risk factors for late atrial fibrillation recurrence were duration of preoperative atrial fibrillation (P =. 01) and Cox maze procedure version (P =. 001). There was no difference in actuarial 10-year survival between the Cox maze procedure versions. The Cox maze procedure remains the gold standard for the treatment of atrial fibrillation and has excellent long-term efficacy. The most significant predictor of late recurrence was duration of preoperative atrial fibrillation, suggesting that earlier surgical intervention would further increase efficacy.