Objective Thyroid ultrasound is crucial for clinical decision in the management of thyroid nodules. In this study, we aimed to estimate and compare the performance of ATA, AACE/ACE/AME and ACR TI-RADS ultrasound classifications in discriminating nodules with high-risk cytology. Design Cross-sectional study. Methods 1077 thyroid nodules undergoing fine-needle aspiration were classified according to ATA, AACE/ACE/AME and ACR TI-RADS ultrasound classifications by an automated algorithm. Odds ratios (ORs) and receiver operating characteristic (ROC) curves for high-risk cytology categories (TIR3b, TIR4 and TIR5) were calculated for the different US categories and compared. Results Cytological categories of risk increased together with all US classifications’ sonographic patterns (P < 0.001). The diagnostic performance (C-index) of ACR TI-RADS and AACE/ACE/AME significantly improved when adding clinical data as gender and age in the regression model (P < 0.001). A significant difference in the final model C-index between the three US classification systems was found (P < 0.029), with the ACR TI-RADS showing the highest nominal C-index value, significantly superior to ATA (P = 0.008), but similar to AACE/ACE/AME (P = 0.287). ATA classification was not able to classify 54 nodules, which showed a significant 7 times higher risk of high-risk cytology than the ‘very low suspicion’ nodules (OR: 7.20 (95% confidence interval: 2.44–21.24), P < 0.001). Conclusions The ACR TI-RADS classification system has the highest area under the ROC curve for the identification of cytological high-risk nodules. ATA classification leaves ‘unclassified’ nodules at relatively high risk of malignancy.