Objective: Patients with dyspepsia with alarm features are suspected of having upper gastrointestinal (GI) malignancy; however, the true value of alarm features in predicting an underlying malignancy for patients with dyspepsia with high background prevalence of Helicobacter pylori infection and upper GI malignancy is uncertain. The aim of the present study was to determine the diagnostic accuracy of alarm features in predicting upper GI malignancy by reviewing an endoscopic database consisting of >100 000 Chinese patients. Methods: A retrospective analysis of prospectively collected data was conducted in a single tertiary medical centre. Consecutive patients who underwent oesophagogastroduodenoscopy (OGD) for dyspepsia in 1996-2006 were enrolled. The data including gender, age, symptoms, and endoscopic and pathological findings were analysed. The main outcome measure was the diagnostic accuracy of individual alarm feature. Results: 102 665 patients were included in the final analysis. Among all the 4362 patients with malignancy, 52% (2258/4362) had alarm features. Among 15 235 patients who had alarm features, 2258 (14.8%) were found to have upper GI malignancy. The pooled sensitivity and specificity of the alarm features were 13.4% and 96.6%, respectively. Only the feature of dysphagia in patients between 36 and 74 years old had a positive likelihood ratio (PLR) >10 for malignancy prediction, while all other alarm features in other age groups had a PLR <10. Conclusions: For uninvestigated Chinese patients with dyspepsia with high background prevalence of H pylori infection and upper GI malignancy, alarm features and age, except for dysphagia in patients between 36 and 74 years old, had limited predictive value for a potential malignancy; therefore, prompt endoscopy may be recommended for these patients. However, less invasive, inexpensive screening methods with high diagnostic yield are still needed to reduce unnecessary endoscopy workload.