To improve the performance of genetic programming algorithm, the niche technology used in genetic algorithm is applied to genetic programming. It is improvement of genetic programming algorithm, which is called NGP in the next text. First, the algorithm fits the data with the original training set. Second, it tracks the extreme points of the fitting function, and according to the dimensions of the fitting function, calculate the extreme points’ Euclidean distance in the independent variable dimension and order it. Then it selects the extreme points whose Euclidean distance is larger and does not exceed the number of the ten percent of the original training set, and added them to the original training set as new training set. Finally, it deals with new training set using genetic programming. In this paper, we use symbolic regression experiment to test the accuracy of the NGP. It illustrates the accuracy and effectiveness of the algorithm.