Estimation and prediction of the extent of surface water resources pollution due to municipal, industrial and agricultural activities, prevention of pollution of these resources and funding of monitoring programs are some of the main issues in water resources management. Many rivers are the main water resource for drinking, agriculture and industry, and variation of their quality is of special importance. Considering the above, estimation and prediction of river water quality parameters along the rivers is one of the main objectives of managers and decision makers in the filed of water resources management. Many models have been established to simulate river water quality. Most of these models need extensive input parameters such as climatological and hydrological data as well as hydraulic input variables such as velocity and cross section of rivers which are difficult or costly to obtain. Qual2e, WASP, and HSPF are some examples of such models. Artificial neural networks (ANN) has been investigated in this research as a prediction tool to estimate river water quality. One of the main advantages of using ANN is their capability to solve nonlinear problems. In this research ANN was used to develop a model which incorporates existing water quality measurements for TDS, EC, BOD, Cl-, pH, HCO3, TH at monitoring stations of Zayande Rud river in order to predict the amount of these parameters in other stations and future time. The input parameters included flow rate, temperature, and precipitation and water quality parameters. Results showed that ANN is capable of predicting the water quality of the river. Considering NMSE, MAE, and R2 as measures of precision of model, ANN performed very well for TDS, EC, and BOD and relatively well for Cl-, pH, HCO3, TH. NMSE and R2 for TDS, EC, and BOD were in the range of 0.62-0.92 and 0.06-0.54, respectively. The prediction model for TDS had the best performance with NMSE of 0.06-0.23 and R2 0.89-0.93. The prediction model for pH had the least desirable performance with NMSE of 0.10-0.45 and R2 0.63-0.90. Results obtained by ANN were compared with those obtained by linear and nonlinear regression models. Comparison of the results clearly shows the superiority of ANN. NMSE and R2 for linear and nonlinear regression models were in the range of 0.42-2.91 and 0.30-0.63, respectively, compared to the values of 0.06-0.39 and 0.68-0.92 for ANN results.