BACKGROUND: Thyroid nodules are relatively common (7% of the population) but are malignant in only 5%-10% of cases. Fine-needle aspiration (FNA) to detect cancer can have > 90% sensitivity but only 50%-65% specificity because of false-positive results, which necessitates surgical controls. We aimed to assess the diagnostic accuracy of immunocytochemistry (ICC) of thyroid FNA to improve its sensitivity and specificity. METHODS: We prospectively collected 2038 thyroid FNAs, of which 1397 were FNA biopsies with liquid-based cytology (Thin-Prep-Hologic(R)). ICC with cytokeratin 19 and HBME1 antibodies (Dako(R) A/S) was used for all malignant cases and cases of atypical cells of undetermined significance (AUS), follicular neoplasm (FN), and nodules suspicious for malignancy-papillary thyroid carcinoma (SM-PTC) as well as some benign cases (abnormal features on radiography or benign on secondary FNA). ICC results were defined as "non-contributory," "favoring benign," "favoring malignant," or "indeterminate." Results for 150 cases were compared with histological controls for diagnostic accuracy. RESULTS: Of these 150 cases ICC was helpful for benign or malignant triage of 48 cases of AUS, FN, and SM-PTC (42% of these lesions). Six (4%) ICC results were false positive (favoring malignant with benign histology) but none were false negative (favoring benign with malignant histology). Results for indeterminate cytological cases favored malignant or benign disease with sensitivity, specificity, and negative and positive predictive values of 100%, 85.2%, 100%, and 86.2%, respectively. CONCLUSIONS: ICC of thyroid FNAs with cytokeratin 19 and HBME1 antibodies can reduce the false-positive and false-negative results of single morphological analyses. It can increase the sensitivity and specificity of diagnosis, thus improving diagnostic accuracy and reducing the need for surgical controls.