BACKGROUND Thyroid fine-needle aspiration (FNA) samples that feature a follicular-patterned, monotonous Hurthle (oncocytic) cell population cannot be diagnosed reliably. The authors of this report recently identified cyclin D3 overexpression on histologic sections of Hurthle cell carcinoma. In this study, they assessed the diagnostic value of cyclin D3 immunohistochemistry added to routine cytology. METHODS Fifty-one FNA samples that were suspicious for Hurtle cell neoplasia and that had histologic follow-up (19 malignant cases) were examined. Cyclin D3 expression levels were evaluated in cell block preparations and were compared with levels of the closely related cyclin D1 protein. RESULTS Greater than 25% positive cells were used as the cutoff point, as suggested by previous studies. Cyclin D1 and cyclin D3 were highly specific (100% for both) and fairly accurate (75% and 92%, respectively) in distinguishing between benign and malignant oncocytic lesions; the positive predictive value (PPV) for each was 100%. However, both cyclins D1 and D3 had low sensitivity (32% and 79%, respectively) and low negative predictive value (NPV) (71% and 89%, respectively). In contrast, by adopting balanced receiver operating characteristic-derived positive cutoff values, cyclin D1 (>or=6.5%) and cyclin D3 (>or=7.5%) were found to be highly sensitive (100% for both) and accurate (90% and 94%, respectively); and the NPV was 100% for both. In contrast, cyclins D1 and D3 had low specificity (84% and 91%, respectively) and a low PPV (79% and 86%, respectively); however, these values improved in samples that were positive for both cyclins (sensitivity, 100%; specificity, 94%; PPV, 90%; NPV, 100%; and accuracy, 96%). CONCLUSIONS Cyclin D3 increased the suspicion of malignancy in indeterminate oncocytic lesions; its diagnostic performance depended on the cutoff point used and was enhanced further when combined with cyclin D1.