In order to evaluate the risk in the process of natural driving, Bayesian network was used to carry out quantitative research on driving risk. First, natural driving test was conducted to obtain the vehicle control data of vehicle spacing, acceleration, and steering entropy, as well as the eye movement data, including the driver’s sight transfer time, saccade speed, and blink frequency. Then, the risk rule of each index data was analyzed, and the risk level threshold of each index was determined. Finally, a vehicle operation risk assessment model based on Bayesian network was constructed, and the risk quantification classification method was proposed. The risk level of vehicle operation was determined, and the sensitivity of the model was analyzed. Results show that the Bayesian network model based on vehicle control and eye movement characterization could effectively conduct quantitative evaluation of driving risk in the driving process. The research results can provide theoretical and technical support for driving risk assessment and early warning in the natural driving process.