Background: Atypia of undetermined significance/follicular lesion of undetermined significance (AUS/FLUS) show variable malignancy risk. We validated the web-based predictive models and the scoring system by the American College of Radiology (ACR) for AUS/FLUS nodules to stratify the malignancy risk. Methods: From January 2011 to November 2014, 275 consecutive nodules diagnosed as AUS/FLUS were enrolled. The discrimination and calibration ability of a web-based scoring with inclusion of biopsy result and sonographic features, old web-based scoring model without biopsy result, and ACR models were assessed. Results: The areas under the receiver operating characteristic curve of the validation set were 0.670 for the new web-based model, 0.710 for the old web-based model, and 0.732 for the ACR scoring risk-stratification model. All models were well calibrated. Conclusion: The web-based scoring risk-stratification model using the combined information of ultrasonography features and biopsy result for AUS/FLUS nodules to stratify malignancy risk presents an acceptable predictive accuracy.