Considering the non-linear, complex and multivariable process of biological denitrification, an activated sludge process was introduced to remove nitrate in groundwater with the aid of artificial neural networks(ANN) to evaluate the nitrate removal effect. The parameters such as COD, NH3-N, NO3--N, NO2--N, MLSS,DO, etc. , were used for input nodes, and COD , NH3 -N , NO3--N , NO2--N were selected for output nodes. Experimental ANN training results show that ANN was able to predict the output water quality parameters very well. Most of relative errors of NO3--N and COD were in the range of ± 10% and ±5% respectively. The results predicted by ANN model of nitrate removal in groundwater produced good agreement with the experimental data. Though ANN model can optimize effect of the whole system, it cannot replace the water treatment process.