This study examines the validity of using ICD-10 codes to identify hospitalized pneumonia cases. Using a case-cohort design, subjects were randomly selected from monthly cohorts of patients aged ⩾65 years discharged from April 2000 to March 2002 from two large tertiary Australian hospitals. Cases had ICD-10-AM codes J10–J18 (pneumonia); the cohort sample was randomly selected from all discharges, frequency matched to cases by month. Codes were validated against three comparators: medical record notation of pneumonia, chest radiograph (CXR) report and both. Notation of pneumonia was determined for 5098/5101 eligible patients, and CXR reports reviewed for 3349/3464 (97%) patients with a CXR. Coding performed best against notation of pneumonia: kappa 0·95, sensitivity 97·8% (95% CI 97·1–98·3), specificity 96·9% (95% CI 96·2–97·5), positive predictive value (PPV) 96·2% (95% CI 95·4–97·0) and negative predictive value (NPV) 98·2% (95% CI 97·6–98·6). When medical record notation of pneumonia is used as the standard, ICD-10 codes are a valid method for retrospective ascertainment of hospitalized pneumonia cases and appear superior to use of complexes of symptoms and signs, or radiology reports.