Because stock market is a complicated, nonlinear and changeable system, prediction of stock price is far more challenging than ordinary time series problems. This paper proposes a novel approach called Dynamic Ordinary Differential Equation (DyODE) modeling of stock market prediction with Gene Expression Programming (GEP). DyODE selects suitable data as training set to build prediction model and hence avoids prediction error caused by using obsolete data. Prediction accuracy of DyODE is tested on the stock prices series. Results show that the accuracy is much higher than that of traditional approaches.