Based on the actual water demands of Beijing city from 2000 to 2011, the forecasting results of BP neural network model, grey GM(1, 1)model, nonlinear model and grey-neural-trend forecasting model and their corresponding model modified by Markov chain were contrasted and analyzed in order to improve the predicting precision of urban water demand. The results showed that the corresponding forecasting model was better than single models, and models modified by Markov chain were better than the unmodified models. In summary, grey-neural-trend forecasting model modified by Markov chain has smaller errors and higher precision accuracy.