Prediction of soil profile water content is important in irrigation, soil and water conservation, and environment improvement, etc. In this paper, we measured daily soil water content at different depths in a red soils field, and estimated the moisture at depth of 20 cm, 30 cm, 40 cm and 60 cm by the time series analysis methods using 10cm measured data . The results showed that there were significant relationships between soil water content at different depths. Based on the water content at depth of 10cm, soil water content at lower layers were predicted by the distributed lag models. With the increase of soil depth to be forecasted, the lag time in the model increased. The average relative error of soil water content between measured and estimated was less than 6 %, and the maximum relative error was not more than 10%. By combining soil water content in 10cm depth and soil water content in 20 cm, 30 cm, 40 em and 60 cm respectively, we simplified the models to autoregressive distributed lag models. With shorter lag times, the models also got acceptable precisions.