Rainfall runoff modeling and prediction of river discharge is one of the important practices in flood control and management, hydraulic structure design and drought management. The present article aims to simulate daily streamflow in Kasilian watershed using an artificial neural network (ANN) and neuro-fuzzy inference system (ANFIS). The intelligent methods have the high potential for determining the relationship between inputs and output. In this study, the input parameters are rainfall, evaporation and temperature of Sangdeh station and streamflow data of Valikbon station are selected as output during 2003 to 2009. The partial auto-correlation function (PACF) was employed for selecting appropriate input parameters to the ANN and ANFIS models. Among different variables in both models, rainfall and evaporation with 1-day lag time were selected as optimal parameters. Then, the results were evaluated using RMSE, NSH, MAE and Rmod statistical criteria for presenting optimal model. The results showed that ANFIS with bell-shaped function and radius of influence=0.14 and NSH=0.80, RMSE=0.056, MAE=0.11, Rmod=0.81 statistical criteria were found to be superior to the ANN with the similar structure, the Levenberg-Marquardt training algorithm, sigmoid transfer function, 14 neurons in the hidden layer and NSH=0.54, RMSE=0.056, MAE=0.14, Rmod=0.87 in testing stage for rainfall-runoff modeling in Kasilian watershed.