In this paper,decision tree model is adopted to mining the most sensitive factors of bus vehicle fuel consumption,based on realworld CAN(Controller Area Network)data collected from bus vehicle in Ningbo,China.By fitting the distribution of bus fuel consumption rate from actual data with Gamma distribution,"high level fuel consumption" is identified.The decision tree model is calibrated based on 10 independent variable about the road condition and driving behaviour.In consideration of the impact of traffic condition and the number of passengers on the bus,the decision tree model is calibrated in three periods to control separately.The result indicates that bus fuel consumption rate is more relative to the interruption of access and the priority of bus lane during the peak hours,which means it is more sensitive to the smoothness of vehicle driving.And during off-peak period,the density of bus stops in the bus line and the frequency of parking during the driving are more significant factors.The quantitative recommendatory ranges of above variables are proposed according to the CART model result as well,and effective fuel-efficient strategies are formulated accordingly at the end.