Abstarct: Maroon River with the length of 422 km is one the important rivers in the Khuzestan Province, Iran. The old name of this river was Tab which originates from Zagros Mountains. After over 120 km, it reaches to Maroon Reservoir at 19 km north east of Behbahan in the Maroon basin. With respect to the importance of the Maroon Reservoir in the region, two methods of Stochastic and Artificial Neural Network (ANN) were used in this research to predict and estimate of inflow discharge into Maroon Reservoir. Data which used in this study included annual and monthly discharge, precipitation and temperature of Idnak hydrometric station, and also monthly precipitation statistics of Cham Nezam, Behbahan and Kheir Abad stations which were collected from Khuzestan Water and Power Authority (KWPA). In order to model discharge two softwares SAMS 2007 and Qnet 2000 were used for stochastic and artificial neural network methods, respectively. In Stochastic models discharge of 41 years (1968-69 to 2008-2009) of Idnak hydrometric station were introduced to the software. Among the Stochastic models for annual series data, Autoregressive (AR), Autoregressive Moving Average(ARMA) and Contemporaneous Autoregressive Moving Average (CARMA) and for monthly data, univariate Periodic Autoregressive Moving Average (MPARMA) and Multivariate Periodic Autoregressive (MPAR) were fitted to the data. Modeling results showed that from different models which fitted to the discharge data of Idnk hydrometric station, it can introduced CARMA(1,1) model with regard to the Akaike and Schwarz Information Criterion (AIC, SIC) as the best model in the simulation and estimation of discharge. ّFor modeling with model of artificial neural network three cases was considered: 1- The effect of transfer functions in the learning of artificial neural network. 2- The effect of hidden laxyers in the learning of artificial neural network. 3- The effect of number of input data on the learning artificial neural network. The result of this research showed that among of the different models which used for artificial neural network, the model with Gaussian transfer function and 3 hidden laxyers has highest correlation coefficient in the training and verification stages rather than other models. For training and verification stages the correlation coefficients of the model were achieved 0.617 and 0.633, respectively. Finally in order to compare artificial neural network and stochastic models, the data divided to 70 and 30 percent for training and verification stages, respectively. Since stochastic model CARMA(1,1) and neural network with Gaussian transfer function and 3 hidden laxyers from were selected from Stochastic models and artificial neural network, the data introduced to these two models. Results and comparison represents that artificial neural network with the highest correlation coefficient and lowest Root Mean Square Error (RMSE) was known as the best model for simulation and estimation of discharge of Maroon river.