The invention relates to a classified estimation method and device for vehicle insurance and belongs to the technical field of intelligent vehicles. The method comprises the following steps: risk characteristic extraction is performed firstly, risk characteristics of a driver and a driving vehicle are obtained, and then 8-dimensional risk characteristics describing driving behaviors of the driver are defined by combining with vehicle-mounted remote information; then a mathematical model for the risk characteristics and loss is established through logistic regression, model parameters are estimated on the basis of historical data with a maximum likelihood method, and value-at-risk thresholds of different risk ranking groups are determined after the model parameters are obtained; finally, risk characteristics and risk values of a to-be-estimated customer are calculated, and risk ranking is performed according to a risk rank table. According to the method, vehicle risks can be ranked through effective combination of driver information, driving vehicle information and driving behavior characteristics, artificial participation is not needed in the whole process, dependence on personal experience is avoided, and the method is direct, effective and convenient to use.