Objective To evaluate the clinical significance and compare the imaging features according to the Thyroid Image Reporting and Data System ( TIRADS) between atypia of undetermined significance/follicular lesion of undetermined significance ( AUS/ FLUS) subcategories. Design & patients A total of 192 thyroid nodules in 188 patients (mean age: 50·2 ± 11·8 years) that had been initially diagnosed as AUS/ FLUS on US-guided fine needle aspiration ( US- FNA) were included. One cytopathologist retrospectively reviewed the cytology slides, subcategorizing cytology results into AUS and FLUS. A TIRADS category was assigned to each thyroid nodule according to the number of suspicious US features. Clinical, US features and malignancy rates were compared between the two subcategories. Results Of the 192 AUS/ FLUS lesions, 149 (77·6%) were subcategorized as AUS and 43 (22·4%) as FLUS. Of the 192 AUS/ FLUS nodules, 82 (42·7%) were malignant. The malignancy rates between AUS and FLUS subcategories were not significantly different, 45·6% to 32·6%, respectively ( P = 0·127). When applying TIRADS, significant differences were seen in TIRADS category between benign and malignant nodules in the AUS subcategory ( P < 0·001), but not in the FLUS subcategory ( P = 0·414). The malignancy rates in TIRADS categories 3, 4a, 4b, 4c and 5 were 15·4%, 22·2%, 33·3%, 57·1% and 80·0% ( P < 0·001) in AUS nodules and 40·0%, 50·0%, 23·5%, 22·2% and 0·0% ( P = 0·414) in FLUS nodules, respectively. Conclusion Suspicious US features are useful in predicting malignancy among AUS subcategories but not in FLUS subcategories. Subcategorization into AUS and FLUS cytology may be helpful in deciding upon treatment or management of thyroid nodules. [ABSTRACT FROM AUTHOR]