The paper discusses the significations of linear and nonlinear cointegration, and indicates that nonlinear cointegration theory can detect the equilibrium between time series more effectively in the nonlinear system. We use wavelet neural network to construct the nonlinear cointegration function, and give the DFP algorithm for network training. In the end, we give out the analysis of Shanghai and Shenzhen stock markets, and the results show the validity of wavelet neural network in nonlinear cointegra...