Abstract: On the basis of an analysis of the characteristics of subsidence,combined with the principle of the artificial neural network, the precipitation, groundwater yield, drawdown of the previous year and degree of consolidation between 1961 and 1980 in the urban district of Tianjin were taken as training net sample inputs and the subsidence over the 20 years as outputs. The subsidence simulation model was constructed after training of the back-propagation network with the Bayesian method. Then the data of 1981 to 1993 were used to check the model. The results indicate that this method with the artificial neural network is an ideal one to predict subsidence. At last the subsidence until 2010 of the urban district of Tianjin was predicted using different levels of precipitation assurance.