The prediction of soil moisture is significant to the research on agricultural production and water cycles.Artificial neural network is used to approximate the phase space reconstruction of chaotic soil moisture time series and the future soil moisture was then predicted.Results show that this method is easier to be used in practice because it only needs one parameter-soil moisture time series.The comparison between the predicted value and the measured value indicates that the prediction method has a little relative error and better prediction accuracy.The study also demonstrates the utility and efficiency of the method for predicting soil moisture.