Gold price has significant nonlinearity and time-variance with many indeterminate influencing factors. In order to improve the forecast accuracy of gold price, this paper puts forward a gold price forecast model combing projection pursuit with neural network. At first, projection pursuit algorithm is used to screen the influencing factors, and then the influencing factors are used as the input variables of BP neural network to learn. Meanwhile, this paper applies genetic algorithm to optimize BP neural network and build gold price forecast model. At last, the forecast performance of the model is tested through simulation experiment. Experimental results show that the combined model can well describe the variation trend of gold price, simplify the network structure and speed up network convergence, which effectively improves the forecast accuracy and operating efficiency of gold price and provides a new forecast method for gold price.