Abstract： This paper introduces the establishment of a predicting model for normalized traffic accident proneness (NAP) after drinking. Eighteen drivers’ EEGs and traffic accident proneness were measured respectively in different drunken states. Considering the instant complexity and longterm periodicity of EEGs measured from the left frontal lobe, the power gain of δ wave and fuzzy entropy of EEG were invented and calculated. A hybrid SigmaPi neural network was introduced and studied to help building the predict model for NAP from the aspects of both power gain of δ wave and fuzzy entropy of EEG. Experiments proved that the predicted values of NAP  accord with the actual values, with a consistent increase and decrease characteristic. When the amount of the alcohol drunk is less than 50% of the subjective maximum alcohol to drink, the errors were very small, but when the amount of the alcohol drunk is more than 50%, the errors became bigger along with the increase volume of alcohol to drink.