The rainfall-runoff modeling is needed to fill in the data or make the data longer. Some method can be used for forecast rainfall processing or runoff like sacramento or artificial neural network (ann). The ann is one of artificial intelligent that is an artificial representation of human's brain which always try to simulation learning process of its. This model is a black box model, so implementation did not need complect science between many aspects in rainfall-runoff happened process. The case study on the upstream of citarum river basin (saguling dam). The data used are a rainfall data (11 rain station) , inflow and sediment rate of month during 19 years from 1986 up to 2004. Rainfall data is input and inflow rate is target output. This research use sacramento and reduced gradient method. The result for training step sacramento's method the correlation is 81 % and reduced gradient's method the correlation is 99 %. For testing sacramento ‘s method the correlation is 83.22 % and reduced gradient's method alternative 2 with four hidden node gives the correlation is 65.57 %. For the next step especially the artificial neural network method still need improvement so that the artificial neural network can be used for modeling of rainfall runoff process. Keywords : rainfall runoff, sacramento, artificial neural network, hidden node, reduced gradient.