Abstract: Due to the stock price is not only affected by many uncertain factors and the data itself has a high degree of fuzzy non-linear characteristics,so there is always a difficult problem for forecasting.In this paper,we firstly use the genetic algorithm with good nonlinear optimization ability to optimize the initial threshold of BP Neural Network,then construc a stock index forecasting model based on historical stock price information as input variable and daily opening price as output variable.An empirical study of the Shanghai Composite Index (opening index) during the observation period showed that the optimized BP Neural Network can not only achieve convergence faster in training process,but also improve the predictability of the training set and the test set.