Actual rainfall runoff modeling involves two kinds of procedures; identification of the optimum parameters and selection of suitable data. Although many reports described automatic procedures to obtain the optimum parameters of models, few proposed a suitable procedure of data selection. In these papers, a new multiple regression model which enables to identify the optimum parameters of linear runoff and select suitable data for the modeling of linear runoff is described. This method is mathematically based on the robust statistical theory developed in the past decade. The method is referred to as RFMD (robust fixed maximum discharge) method because it is also based on the general expression of Shiraishi’s FMD method of data separation. In this paper, the robust statistical theory was introduced into runoff analysis and tested by using the manipulated data of the Zao dam in Japan. Compared to the FMD method, the RFMD method showed a strong robustness.