From Longley, the various equations for determining the runoff to water management are presented by the researchers that are widely used in hydrologic sciences. In this study by using observational data, was an evaluated empirical, artificial neural network (ANN) model in estimation of runoff coefficient. The study area was Bar Ariyeh Neishabour watershed. The data of 33 flood events during 1952 to 2006 were collected. Among Characteristics from precipitation hytographs as model input variables were extracted include The average intensity of rainfall, average rainfall, 1 to 4rd quartiles of rainfall, 1 to 4rd quartiles intensity of rainfall, total precipitation of five days before, the index ϕ. Therefore, using these parameters and different combinations in the input layer network, different networks were implemented. Artificial neural network is used learning algorithm with Levenberg-Marqwart and Hyperbolic tangant trained and performed with various inputs. The results showed, network with 1 to 4rd quartiles intensity of rainfall, average rainfall, time of rainfall, total precipitation of five days before and ϕ index as input layer with Hyperbolic tangant transfer function could predict storm runoff coefficient with determination coefficient 0.98 and the Root Mean Squared Error 0.0337 and Mean Absolute Error 0.0275.