Abstract: Following wavelet packet transformation, the research on stock market forecast is conducted and a wavelet packet method is proposed. First, using wavelet packet transformation, share index time series are decoded into several parts. And, to each part, its chaos features are analyzed. Then, chaotic forecasting models are established to forecast each part respectively. Finally, the forecasting results of chaotic models are reconstructed based on the wavelet packet theory and the forecasting result of the original time series can be obtained. According to this method, the research on one step forecast and multi-step forecast of Shanghai Composite Index daily gain rate series is completed and the results are well documented.