Objective Prospective data on the accuracy of ultrasound (US) classification systems in thyroid nodules are still scarce. The aim of this study is to compare the accuracy of the American College of Radiology Thyroid Imaging Reporting and Data System (ACR‐TIRADS) and European (EU)‐TIRADS classification systems. Design and Patients Consecutive patients with one or more thyroid nodule(s) who underwent fine‐needle aspiration (FNA) under ultrasonographic guidance (FNA‐US) were prospectively evaluated. Measurements Clinical evaluation and US data were collected. The reference standard used for this study was FNA‐US cytology and histopathological diagnosis. Results A total of 186 thyroid nodules in 166 patients were evaluated, resulting in 168 nodules from 149 patients with conclusive benign or malignant results. Sensitivity, specificity, negative predictive value (NPV) and false negative (FN) were 100.0%, 28.7%, 100.0% and 0.0%, respectively, for ACR‐TIRADS; and 90.0%, 19.1%, 96.8% and 9.1% (n = 1), respectively, for EU‐TIRADS. The number of unnecessary FNA‐US indicated by ACR‐TIRADS was lower than EU‐TIRADS (71.3% vs. 80.9%, p = .017), and the number of possibly avoided FNA‐US was higher (26.7% vs. 17.8%). Using the same threshold of ACR‐TIRADS to indicate FNA‐US in EU‐TIRADS 3 nodules (2.5 cm), there was an improvement in specificity (30.6%) and avoided FNA‐US (28.6%). The best performance of both systems was demonstrated when FNA‐US would be indicated only in highly suspicious nodules and/or in the presence of lymphadenopathy, with 85.7% and 89.3% of possibly avoided FNA‐US for ACR‐TIRADS and EU‐TIRADS, respectively, without increasing FN. Conclusion Both systems presented high sensitivity, but low specificity in selecting nodules for FNA‐US. The use of nodular size for FNA‐US selection is questioned.