The Bethesda System for Reporting Thyroid Cytopathology previously described 4subclassesof atypia within the Atypia of Undetermined Significance (AUS) category: nuclear (AUS-Nuc), architectural (AUS-A), oncocytic (AUS-Onc), and atypia not otherwise specified (AUS-NOS). Accumulating evidence supports a binary AUS subclassification scheme based primarily on the presence of nuclear atypia only. The purpose of this study is to compare the risk stratification of binary versus 4-tier AUS subclassification systems among AUS nodules with molecular and/or histologic follow-up. Thyroid aspirates classified as AUS and tested using Afirma (Veracyte, Inc.) between 6/2013 and 7/2021 were included. For resected nodules, histological classification was considered as the final outcome. For unresected nodules, benign Afirma results were considered low-risk outcomes, similar to histologically benign nodules. Suspicious or nondiagnostic Afirma results were considered indeterminate outcomes. The prevalence of outcomes warranting surgery (noninvasive follicular thyroid neoplasm with papillary-like nuclear features [NIFTP] or cancer) was calculated for each AUS subclass. A total of 559 AUS nodules with Afirma testing were identified. Excluding nodules with indeterminate molecular outcomes, NIFTP/cancer prevalence for AUS-Nuc was 21% (57/266), which was higher than that for AUS-A (6%, 11/188), AUS-Onc (8%, 4/53), and AUS-NOS (0%, 0/9). A binary AUS subclassification scheme based on nuclear atypia showed a significant difference in NIFTP/cancer prevalence (21% versus 6%, P < 0.0001). Binary reporting of AUS subclasses based on nuclear atypia distinguishes cases with a higher risk of NIFTP/cancer. There is a low but non-negligible prevalence of NIFTP/cancer in cases without nuclear atypia.