Significant interobserver variabilities exist for Bethesda category III: atypia of undetermined significance (AUS) of The Bethesda System for Reporting Thyroid Cytopathology (TBSRTC). Thus, subcategorization of AUS including AUS “nuclear” and AUS “other” is proposed in the recent 3rd edition of TBSRTC. This study investigated the impact of the nuclear features/architectural features/nuclear score (NS) (3-tiered)/subcategories and subgroups on risk of malignancy (ROM) in thyroid fine-needle aspirations (FNA). 6940 FNAs were evaluated. 1224 (17.6%) cases diagnosed as AUS were reviewed, and 240 patients (initial FNAs of 260 nodules and 240 thyroidectomies) were included. Subcategories and subgroups were defined according to TBSRTC 2nd and 3rd editions. Histological diagnostic groups included nonneoplastic disease, benign neoplasm, low-risk neoplasm, and malignant neoplasm. Overall, ROM was 30.7%. ROM was significantly higher in FNAs with nuclear overlapping (35.5%), nuclear molding (56.9%), irregular contours (42.1%), nuclear grooves (74.1%), chromatin clearing (49.4%), and chromatin margination (57.7%), and these features were independent significant predictors for malignancy. FNAs with NS3 had significantly higher ROM (64.2%). Three-dimensional groups were significantly more frequent in malignant neoplasms (35.7%). ROM was significantly higher in AUS-nuclear subcategory (48.2%) and in AUS-nuclear and architectural subcategory (38.3%). The highest ROM was detected in AUS-nuclear1 subgroup (65.2%). ROM was significantly higher in the group including AUS-nuclear and AUS-nuclear and architectural subcategories, namely “high-risk group” than the group including other subcategories, namely “low-risk group” (42.0%vs 13.9%). In conclusion, subcategorization may not be the end point, and nuclear scoring and evaluation of architectural patterns according to strict criteria may provide data for remodeling of TBSRTC categories.