BACKGROUND Hepatocellular carcinoma (HCC) is routinely graded histopathologically using a modified Edmondson system (ES). The cytologic grading of HCC has been used predominantly as an adjunct in differential diagnoses (i.e., to help distinguish HCC from other tumors as well as nonneoplastic lesions). However, there are unanswered questions regarding the reliability of the cytologic grading of HCC and its correlation with histologic follow-up. METHODS A total of 106 cases of HCC were identified in the authors' cytopathology files from 1977 to the present. Of these cases, 64 had either a core needle or excisional biopsy sample that was judged to be adequate for histologic grading. From each case smears were graded independently in a blinded fashion by two cytopathologists, and tissue slides were graded by a liver pathologist. The cytopathologists' grading was then adjudicated by considering the histologic diagnosis as the "truth standard". Finally, after the scores were calculated, a statistical analysis was performed to ascertain the accuracy of the cytopathologic grading. RESULTS The sensitivity for accurate grading was found to be highest for well differentiated (WD) lesions; the specificity was found to be highest for poorly differentiated (PD) HCC for both cytopathologists. Interobserver agreement was highest for WD HCC. WD HCC displayed cohesive fragments, often associated with characteristic vascular/endothelial patterns. In addition, moderately differentiated (MD) HCC demonstrated numerous single cells and atypical naked nuclei, usually with prominent nucleoli. PD HCC displayed loose nests and three-dimensional fragments (often gland-like), pleomorphism, macronucleoli, and focal necrosis. CONCLUSIONS In the authors' experience, the three-tier cytologic grading of HCC was found to be only moderately accurate. The accuracy of cytologic grading was reported to be high for WD/PD HCC and low for MD HCC. The architectural criteria appear to be more useful for WD HCC, whereas marked cellular pleomorphism is specific for PD HCC. The authors propose that a two-tier grading system may be more useful, given the recent studies of HCC recurrence.