BACKGROUND: The practice of onsite cytology interpretation varies across endoscopic ultrasound (EUS) programs in the Untied States and Europe. The value, reliability, and agreement of rapid onsite evaluation (ROSE) compared with final interpretation and its impact on patient management remain largely unknown. We compared agreement between ROSE of EUS-FNA (endoscopic ultrasound-guided fine needle aspiration) specimens with final cytology interpretation and their respective operating characteristics. METHODS: We prospectively evaluated consecutive EUS-FNA specimens obtained by a single endosonographer in the presence of an attending cytopathologist (July 2000-November 2002). The "agreement" between ROSE and final cytology interpretation was assessed by the "kappa" statistic. The frequency and possible reasons for discrepancy between ROSE and final cytologic interpretation were determined. RESULTS: A total of 540 patients (median age 63 yr, 77% white) underwent EUS-FNAs of 656 lesions. These included lymph nodes (LNs, N = 248), solid pancreatic masses (SPMs, N = 229), cystic pancreatic masses (CPM, N = 57), mural lesions (41), bile duct/gallbladder (N = 28), liver (N = 17), mediastinum/lung (N = 17), adrenal (N = 15), spleen (N = 3), and kidney (N = 1). Data were available for onsite evaluation in 607 lesions. Intitial cytology was benign (N = 243), atypical (N = 23), suspicious (24), malignant (300), and indeterminate (N = 17). Out of the 243 benign lesions interpreted by onsite evaluation, five lesions (2.1%) were upgraded to be malignant or suspicious for malignancy on final cytology report. Of the 300 lesions initially reported malignant on ROSE, 294 (98%) remained malignant on the final cytology. There was an excellent agreement between ROSE and final cytologic evaluation (kappa = 84.0%, 95% CI 80.2-87.7), Compared with the true final status, accuracy for final interpretation was slightly higher than for ROSE but was not statistically significant (95.8% vs 93.9%). Scant cellularity remained the most frequent reason for discrepancy, accounting for 21 of 51 discrepancies (41%). Other reasons for discrepancy included: cases requiring an intradepartmental consultation (22%), cases requiring ancillary studies (12%), intra-observer variability (10%), and challenging diagnosis (10%). In three cases, (6%) we were unable to determine the possible reason for discrepancy. CONCLUSION: ROSE of EUS-FNA specimens is highly accurate compared with final cytologic evaluation. Because the diagnosis of malignancy rarely changes, informing our patients and their families and our referring physicians of a cancer diagnosis significantly reduces physician work load and expedites patient management. The majority of discrepancies are related to the nature of the lesion either because it sheds few cells, has challenging morphology, and/or requires additional ancillary studies. © 2006 by Am. Coll. of Gastroenterology.