Although many large series demonstrate the effectiveness of thyroid fine-needle aspiration (FNA), measuring its accuracy has been suboptimal owing to inappropriate statistical methods. All thyroid fine-needle aspirates were correlated with corresponding histologic and cytologic follow-up for a 4-year period, and the accuracy was determined using receiver operator characteristic curves, which allow inclusion of nondiagnostic and indeterminate cases. There were 1,085 cases, 291 with follow-up. The overall accuracy was 0.90 +/- 0.02 for a single aspiration session. A nondiagnostic aspirate was associated with a significant risk of malignancy (16%). However, 70% of patients who underwent reaspiration had adequate and negative results, and reaspiration significantly increased overall accuracy. Subcategorizing the nondiagnostic category did not affect accuracy, but did define categories with a significantly different change of a negative diagnosis on repeated aspiration. Although subcategories of papillary carcinoma were associated with significantly different risks of carcinoma (40% vs 81%), they did not significantly improve overall accuracy. Receiver operator characteristic curves can be used to define the accuracy of thyroid FNA. This method demonstrates significantly increased accuracy with repeated aspiration of nondiagnostic cases and demonstrates that subcategorization does not improve the overall accuracy of the test.