An improved gene expression programming (GEP) algorithm is presented, the algorithm has adaptive and better astringent capability, and is applied to function optimization. In the evolution process the mutation probability and crossover probability will be increased adaptively with the increase of fitness. The feedback of fitness value to calculate the mutation probability and crossover probability, this step is added to the algorithm. Put up linear regress experiments to the improved algorithm and get a good effect. Experiments result shows that the algorithm has very good capability in function optimization.