The constraints of automatic test paper are optimized, and the algorithm is improved according to the examination hierarchy of knowledge points . The knowledge points weight grading table is used when initializing the population , and the chromosomes in the initial population are selected according to the knowledge weight order so that the population can meet the requirement of the assessment level of the knowledge points from the initial time. In the design of the genetic operators maintains the stability of the population of knowledge point .The results show that the improved algorithm can improve the coverage rate of important knowledge points while guaranteeing the coverage of knowledge points.