Influenza causes severe illness and deaths, and global surveillance systems use different clinical case definitions to identify patients for diagnostic testing. We used data collected during January 2007–July 2010 at hospital-based influenza surveillance sites in western Kenya to calculate sensitivity, specificity, positive predictive value, and negative predictive value for eight clinical sign/symptom combinations in hospitalized patients with acute respiratory illnesses, including severe acute respiratory illness (SARI) (persons aged 2–59 months: cough or difficulty breathing with an elevated respiratory rate or a danger sign; persons aged ⩾5 years: temperature ⩾38 °C, difficulty breathing, and cough or sore throat) and influenza-like illness (ILI) (all ages: temperature ⩾38 °C and cough or sore throat). Overall, 4800 persons aged ⩾2 months were tested for influenza; 416 (9%) had laboratory-confirmed influenza infections. The symptom combination of cough with fever (subjective or measured ⩾38 °C) had high sensitivity [87·0%, 95% confidence interval (CI) 83·3–88·9], and ILI had high specificity (70·0%, 95% CI 68·6–71·3). The case definition combining cough and any fever is a simple, sensitive case definition for influenza in hospitalized persons of all age groups, whereas the ILI case definition is the most specific. The SARI case definition did not maximize sensitivity or specificity.