Patients with abdominal pain and no definite diagnosis referred for endoscopy were studied to define discriminating features in the history, and the value of a stool occult blood test, in predicting the presence of upper gastrointestinal disease. Endoscopy was performed in 116 patients; pathology was seen in 32 (duodenal ulcer 17, gastric carcinoma 4, gastric ulcer 3, miscellaneous 8) and no pathology was seen in 84 patients. Features that predicted upper gastrointestinal pathology were, in descending order of rank: a positive pointing sign, a positive stool Fecult test, a history of vomiting, loss of weight, and alcohol intake. Using these discriminating features together it was possible to correctly predict 95% of patients with abnormal endoscopy and 82% of patients with a normal endoscopy. The history and the stool occult blood test are useful predictors of the presence of upper gastrointestinal pathology and may aid rational selection of patients for endoscopy.