Our objectives were to devise a cytologic grading system and determine whether it would predict survival of patients with solid-type pancreatic adenocarcinoma. We evaluated 116 consecutive patients from July 2000 to November 2002; they were followed up until September 2003. We scored the following features on rapid Romanowsky-stained endoscopic ultrasound-guided fine-needle aspiration smears: cell group architecture, single cells, nuclear grade, mucus, bizarre cells, and necrosis. A cytologic grade (low vs high) was assigned. The Kaplan-Meier estimate of 6-month survival was 76% (SE, 7%) for patients with low-grade tumors vs 50% (SE, 6%) for patients with high-grade carcinoma. The median survival for patients with low-grade vs high-grade tumors was 1 year vs 6 months, respectively (chi2 = 4.45; P = .035). Cox proportional hazards regression showed tumor stage, cancer-specific treatment, and cytologic grade to be independent predictors of survival (P = .001). No other factors (age, mass location, placement of stent, presence of concomitant chronic pancreatitis, race, sex) predicted survival. We devised a grading system that independently predicted survival in patients with pancreatic adenocarcinoma.