Objective To establish a prediction model for recurrent laryngeal nerve thermal injury (RLNTI) in radiofrequency ablation (RFA) of papillary thyroid carcinoma (PTC). Materials and methods This prospective study was conducted at a single center from July 2016 to September 2022. Patients diagnosed with PTC and received US-RFA by different interventional therapy groups were enrolled, and classified as derivation cohort and validation cohort based on therapy groups. The logistic regression analysis was used in derivation cohort to develop the model and internal and external validation was performed in derivation and validation cohort respectively. Results The derivation cohort included 1632 patients (mean age, 40 years, +/- 11 (standard deviation), 1300 female), and the validation cohort comprised 755 patients (mean age, 38 years, +/- 11 (standard deviation), 584 female). Three variables were independently associated with RLNTI: the minimum distance from tumor to tracheoesophageal groove (TEG-D) <= 2.95 mm (odds ratio, 8.179; p < 0.0001), tumor posterior location (odds ratio, 3.849; p = 0.0008) and ablation energy > 0.695 kcal (odds ratio, 4.537; p < 0.0001). The area under the receiver operating characteristic curve (AUC) of the model in the derivation cohort, validation cohort, and total group were 0.877 (95% CI: 0.845, 0.910), 0.901 (95% CI: 0.835, 0.967) and 0.886 (95% CI: 0.85, 0.915) respectively. The calibration plot and Hosmer-Lemeshow test demonstrated a good fit (chi(2) = 3.49, p = 0.321). Conclusion The prediction model of RLNTI was derived based on TEG-D, tumor posterior location, and ablation energy and exhibited good discrimination and calibration abilities.