Abstract: A back propagation (BP) artificial neural network model was set up to predict the effect of nitrogen removal using a surface water purification device. The observed data of water quality parameters were used as study sample, and the raw water TN, ammonium nitrogen, nitrate nitrogen, COD_Mn and operation time of the device selected as projection parameter in this model. Besides, the multivariate linear regression model was introduced to compare with BP neural network. The results showed that the coefficient of determination of BP artificial neural network model was 0.985, which stayed at a high level. And the maximum error was 5.92%. Obviously, BP artificial neural network model was more precise, faster and better than multivariate linear regression model. It could accurately predict the removal effect of TN by purification device.