Abstract Soil Moisture Accounting(SMA) process is an important topic in rainfall-runoff analysis because it determines the initial dis-charge in hydrological models. The SMA process is used in continuous NRCS method. However, it can not explain some eventsdue to its assumption of initial rainfall losses. In this study, a modified SMA procedure is applied to solve the problem with theinitial rainfall losses assumption of the NRCS method, and a modified GR4J model is developed using the modified SMA pro-cedure. To verify the applicability of the proposed model, the results are compared with those from existing GR4J models in threedifferent Korean basins. In this process, Harmony Search optimization algorithm is used to calibrate model parameters. And thecalibrated parameters are verified to check the representability of the target basin. The developed model shows better results thanexisting models do and can improve accuracy of the rainfall-runoff model. Since the modified GR4J model shows outstandingefficiency while using less parameters, it is expected to be applied to many other Korean basins.Key words : Soil moisture accounting, Parameter optimization, GR4J, Long term rainfall-runoff analysis