Abstarct: Following the results of previous researches on the neural network modeling application in order to predict the wastewater treatment plants behavior and performance, in this research a multi-laxyered artificial neural network of perceptron to evaluate the wastewater treatment plant performance of the Birjand city, which is in the form of the stabilization pond, has been employed. In this regard, baxsed on the monthly measurements done at the Birjand wastewater treatment plant, the target variables in the effluent of the maturation pond were predicted in various scenarios. The input variables of the neural network are include temperature, biological oxygen content (BOD), chemical oxygen demand (COD), concentration of total suspended solids (TSS) and the amount of wastewater discharge entering the anaerobic pond and the network output variables are include BOD, COD, TSS of the effluent from maturation pond, for 131 months from August 2006 to June 2017. Thus, various artificial neural networks were designed and compared in terms of training algorithms, number of output variables, number of hidden laxyer neurons and combination of activation functions. The results showed that the best algorithm for network training is the LM algorithm. It was also found that the optimal arrangement of neurons for BOD, COD and TSS variables was 5-19-1, 5-19-1, and 5-20-1, respectively, with correlation coefficients of 0.90, 0.96 and 0.95. The optimal combination of activation functions for these variables showed that sigmoid functions have better performance in predicting the behavior of variables. On the other hand, according to the obtained results, it can be concluded that the single-output networks have a better result than others. baxsed on the sensitivity analysis of modified weights of optimized neural network, it was determined that the most effective input parameters on the output variables of the BOD, COD and TSS are TSS, BOD, and the temperature of the wastewater entering the anaerobic pond. As a result, by comparing the results of prediction with other researches and according to statistical criteria, we can ensure the efficiency of the neural network for modeling the Birjand wastewater treatment plant.