Using clinical predictors, we evaluated clinical case definitions of influenza during the 1995-1996 outbreak in France. Thirty-five general practitioners collected virological specimens and clinical data. Predictors of influenza virus infection were selected with logistic regression models. The results varied with the influenza virus subtype: temperature of >38.2 degrees C, stiffness or myalgia, rhinorrhea, and cough were predictive of influenza A/H3N2, whereas fatigue, lacrimation or conjunctival injection, and the absence of stiffness or myalgia were predictive of influenza A/H1N1. On the basis of this analysis and data from the literature, 12 clinical case definitions were evaluated for their abilities to diagnose influenza virus infection. They were associated with positive predictive values of 27% to 40% and negative predictive values of 80% to 91%. We conclude that focused studies evaluating clinical case definitions of influenza with use of subsets of patients should accompany population-based disease surveillance for optimal estimates of the disease burden associated with influenza epidemics.