Background: Decisions about multimodality treatment for upper gastrointestinal malignancies are largely predicted on clinical staging information. However, hospital-level accuracy of clinical staging is currently unknown. Methods: A national cohort study of patients with adenocarcinoma of the esophagus, stomach, or pancreas in the NCDB (2006-2015) who were treated with upfront resection. Hospital-level staging accuracy (ascertained by comparing clinical stage to pathologic stage) was calculated. Within hospital correlation of staging accuracy across disease sites was evaluated using risk and reliability adjustment. Results: Overall, 1246 hospitals were evaluated. Median hospital T-staging accuracy was 77.5%, 73.7%, and 60.8% for esophageal, gastric, and pancreatic cancer, respectively. Median hospital N-staging accuracy was 80.2%, 72.9%, and 61.8%, respectively. For T-stage, over-staging was most frequently observed in esophageal patients (11.2%) while under-staging was most frequent in pancreatic patients (36.1%). For N-stage, over-staging was infrequent for all three, while under-staging was most common in pancreatic patients (37.4%). Correlation across disease sites was weak for both T-(best observed, r = .34) and N-stages (r = .30). When high volume hospitals were evaluated, correlation improved but accuracy rates were similar. Conclusions: Despite the importance of clinical staging in multimodality treatment planning, hospitals inaccurately stage 20-40% of patients, with low correlation across disease sites.