Abstract: A novel collision risk inference method based on case scenario inversion is introduced to study the evolution of the accident. A track calculation and two-dimensional interpolation algorithm is developed to invert the overtaking process and the development of collision scenario according to the dynamic information collected during the whole process of the accident. The virtually repeated process is analyzed and the evolution of collision criteria is examined. A negative exponential function is introduced to calculate the I_CR（Collision Risk Index） that comprehensively indicates the degree of collision danger. A dimensionless collision risk index is defined to comprehensively represent the status of d_CPA（Distance of Closest Point of Approach）, t_CPA（Time to Closest Point of Approach） and the parameters of both ships. A grey cloud inference model is developed. 28 samples of ship overtaking accidents are examined with the model. The following findings are interesting: in the course of ship overtaking, an accident, if happens, will develop in three stages, that is, collision risk, close situation and immediate danger; the time duration of each stage are almost the same for all the accidents examined; the P_CR（Potential Collision Risks） for the three stages are 0.349 1, 0.557 5 and 0.777 1 in average respectively.