ObjectiveThyroid Imaging Reporting and Data System (TIRADS) is a simple and reliable reporting system for the prediction of malignancy. We aimed to determine the role of TIRADS in the prediction of malignancy in subcategories of Bethesda Category III, atypia of undetermined significance (AUS) and follicular lesion of undetermined significance (FLUS).,Design & patientsA total of 461 nodules with AUS cytology in 450 patients and 179 nodules with FLUS cytology in 168 patients were included. Ultrasonography (US) features and postoperative histopathology results were documented. Every suspicious US feature was scored as 1 and 0 according to the presence or not, respectively. TIRADS category of each nodule was determined.,ResultsIn AUS subcategory, histopathologically malignant nodules had significantly different TIRADS categories compared to benign nodules (P = 0001), but this was not the case in FLUS subcategory (P = 0121). In AUS group, malignant nodules had significantly higher prevalance of microcalcification, hypoechogenicity and anteroposterior/transverse ratio than benign ones (P < 0001, P < 0001 and P = 0003, respectively) and TIRADS categories of 4c and 5 were more frequent in malignant nodules (P < 005). Microcalcification, hypoechogenicity and TIRADS were found to be associated with malignancy in multivariate logistic regression analysis in this subcategory. TIRADS category 4c was associated with malignancy (AUC SE: 0584 +/- 0028). In FLUS subcategory, there was no significant difference between histopathologically malignant and benign nodules with respect to suspicious US features (P > 005, all).,ConclusionTIRADS seems to be useful in predicting malignancy and planning further management in the AUS subcategory, but not quite so in the FLUS subcategory.