With regard to the characteristic diversity of subjects, updated flexible contents, and unified examination papers for multiple courses in Unified National Graduate Entrance Examination, a random examination questions extraction algorithm based on multi-layer filtering model and dynamic probability model is proposed. First, the multi-layer filtering model is used to filter out the examination questions that meet the requirements for the examination, and then the dynamic probability model is used to adjust all the examination questions dynamically according to a certain probability. The results show that a reasonable distribution of questions has been achieved, which can well meets the requirements for the examination questions extraction in the National Graduate Entrance Examination.